A Database of Information on the 6,355 Kanji in the JIS X 0208 Standard

Copyright (C) 2001 The Electronic Dictionary Research and Development Group, Monash University.

(NB: This document has been converted quickly from plain text to HTML. As a result, some of the formatting has been left as it was in the original document. A more elegant version may be developed later.)



The KANJIDIC file contains comprehensive information about Japanese kanji. It is a text file currently 6,355 lines long, with one line for each kanji in the two levels of the characters specified in the JIS X 0208-1990 set. (For information about this set, see Appendix A.)

The file contains a mixture of ASCII characters and kana/kanji encoded using the EUC (Extended Unix Code) coding.

Attention is drawn to the KANJIDIC LICENCE STATEMENT AND COPYRIGHT NOTICE included below in this document.

A similar file, KANJD212, is available for the 5,801 supplementary kanji in the JIS X 0212-1990 set.


The first part of each line is of a fixed format, indicating which character the line is for, while the rest is more free-format.

The first two bytes are the kanji itself. There is then a space, the 4-byte ASCII representation of the hexadecimal coding of the two-byte JIS encoding, and another space.

The rest of the line is composed of a combination of three kinds of fields (which may be in any order and interspersed):

  1. information fields, beginning with an identifying letter and ending with a space. See below for more information about these fields.

  2. readings (with '-' to indicate prefixes/suffixes, and '.' to separate a reading from its okurigana). ON-yomi are in katakana and KUN-yomi are in hiragana. There may be several classes of reading fields, with ordinary readings first, followed by members of the other classes, if any. The current other classes, and their tagging, are:
    1. where the kanji has special "nanori" (i.e. name) readings, these are preceded the marker "T1";

    2. where the kanji is a radical, and the radical name is not already a reading, the radical name is preceded the marker "T2".

      (Other Tn classes may be created at a later date.)

  3. English meanings. Each such field begins with an open brace '{' and ends at the next close brace '}'.

There are currently a variety of predefined fields (programs using KANJIDIC should not make any assumptions about the presence or absence of any of these fields, as KANJIDIC is certain to be extended in the future):

B<num> -- the radical (Bushu) number. There is one per entry. As far as possible, this is the radical number used in the Nelson "New Japanese-English Character Dictionary. Where the classical or historical radical number differs from this, it is present as a separate C<num> entry.

C<num> -- the historical or classical radical number, as recorded in the KangXi Zidian (where this differs from the B<num> entry.) There will be at most one of these.

F<num> -- the frequency-of-use ranking. At most one per line. The 2,135 most-used characters have a ranking; those characters that lack this field are not ranked. The frequency is a number from 1 to 2,135 that expresses the relative frequency of occurrence of a character in modern Japanese. The data is based on statistics published by The National Language Research Institute (Tokyo), interpreted and adapted by Jack Halpern in a manner to make it useful to the learner. The data is derived from the New Japanese-English Character Dictionary (Kenkyusha, Tokyo 1990; NTC, Chicago 1993). The commercial utilization of the frequency numbers is prohibited without written permission from Jack Halpern. Use by individuals and small groups for reference and research purposes is permitted, on condition that acknowledgment of the source and this notice are included.

G<num> -- the Jouyou grade level. At most one per line. G1 through G6 indicate Jouyou grades 1-6. G8 indicates general-use characters. G9 indicates Jinmeiyou ("for use in names") characters. If not present, it is a kanji outside these categories.

H<num> -- the index number in the New Japanese-English Character Dictionary, edited by Jack Halpern. At most one allowed per line. If not preset, the character is not in Halpern.

N<num> -- the index number in the Modern Reader's Japanese-English Character Dictionary, edited by Andrew Nelson. At most one allowed per line. If not present, the character is not in Nelson, or is considered to be a non-standard version, in which case it may have a cross-reference code in the form: XNnnnn. (Note that many kanji currently used are what Nelson described as "non-standard" forms or glyphs.)

V<num> -- the index number in The New Nelson Japanese-English Character Dictionary, edited by John Haig.

D<code> -- the "D" codes will be progressively used for dictionary based codes.

(a) DRnnnn - these are the codes developed by Father Joseph De Roo, and published in his book "2001 Kanji" (Bojinsha). Fr De Roo has given his permission for these codes to be included.

(b) DKnnnn - the index numbers used by Jack Halpern in his Kanji Learners Dictionary, published by Kodansha in 1999. The numbers have been provided by Mr Halpern.

(c) DOnnnn - the index numbers used in P.G. O'Neill's Essential Kanji (ISBN 0-8348-0222-8). The numbers have been provided by Glenn Rosenthal.

P<code> -- the SKIP pattern code. The <code> is of the form "P<num>-<num>-<num>". The System of Kanji Indexing by Patterns (SKIP) is a scheme for the classification and rapid retrieval of Chinese characters on the basis of geometrical patterns. Developed by Jack Halpern, it first appeared in the New Japanese-English Character Dictionary (Kenkyusha, Tokyo 1990; NTC, Chicago 1993), and is being used in a series of dictionaries and learning tools called KIT (Kanji Integrated Tools). SKIP is protected by copyright, copyleft and patent laws. The commercial utilization of SKIP in any form is strictly forbidden without the written permission of Jack Halpern, the copyright holder (jhalpern@cc.win.or.jp). (A brief summary of the method is in Appendix C. See Appendix E. for some of the rules applied when counting strokes in some of the radicals.)

S<num> -- the stroke count. At least one per line. If more than one, the first is considered the accepted count, while subsequent ones are common miscounts. (See Appendix E. For some of the rules applied when counting strokes in some of the radicals.)

U<hexnum> -- the Unicode encoding of the kanji. See Appendix B for further information on this code. There is exactly one per line.

I<code> -- the index codes in the reference books by Spahn & Hadamitzky. These codes take two forms:

(a) for The Kanji Dictionary (Tuttle 1996), they are in the form nxnn.n, e.g. 3k11.2, where the kanji has 3 strokes in the identifying radical, it is radical "k" in the S&H classification system, there are 11 other strokes, and it is the 2nd kanji in the 3k11 sequence. I am very grateful to Mark Spahn for providing the (almost) full list of these descriptor codes for the kanji in this file. At the time of writing some 800 kanji in the file lack the SH descriptor. This is because the book used a different glyph as the primary kanji. The gaps are gradually being filled in. Where the JIS X 0208 glyph is the second kanji for a particular descriptor code, it has a "-2" appended to the code.

(b) for the Kanji & Kana book (Tuttle), they are in the form INnnnn, where nnnn is the number of the 1,945 kanji referenced in that book.

Qnnnn.n -- the "Four Corner" code for that kanji. This is a code invented by Wang Chen in 1928, it has since then been widely used for dictionaries in China and Japan. In some cases there are two of these codes, as it is can be little ambiguous, and Morohashi has some kanji coded differently from their traditional Chinese codes. See Appendix D for an overview of the Four Corner System. Christian Wittern, who passed on these codes, comments that they are in need of proof-reading and thus users are advised to be cautious using the codes for serious scholarship.

MNnnnnnnn and MPnn.nnnn -- the index number and volume.page respectively of the kanji in the 13-volume Morohashi "Daikanwajiten. In the MNnnn field, a terminal `P`, e.g. MN4879P, indicates that it is 4879' in the original. In some 500 cases, the number is terminated with an `X`, to indicate that the kanji in Morohashi has a close, but not identical, glyph to the form in the JIS X 0208 standard.

Ennnn -- the index number used in "A Guide To Remembering Japanese Characters" by Kenneth G. Henshall. There are 1945 kanji with these numbers (i.e. the Jouyou subset.)

Knnnn -- the index number in the Gakken Kanji Dictionary ("A New Dictionary of Kanji Usage"). Some of the numbers relate to the list at the back of the book, jouyou kanji not contained in the dictionary, and various historical tables at the end.

Lnnnn -- the index number used in "Remembering The Kanji" by James Heisig.

Onnnn -- the index number in "Japanese Names", by P.G. O'Neill. (Weatherhill, 1972) (A warning: some of the numbers end with 'A'. This is how they appear in the book; it is not a problem with the file.)

Wxxxx -- the romanized form of the Korean reading(s) of the kanji. Most of these kanji have one Korean reading, a few have two or more. The readings are in the (Republic of Korea) Ministry of Education style of romanization.

Yxxxxx -- the "Pinyin" of each kanji, i.e. the (Mandarin or Beijing) Chinese romanization. About 6,000 of the kanji have these. Obviously most of the native Japanese kokuji do not have Pinyin, however at least one does as it was taken into Chinese at a later date.

Xxxxxxx -- a cross-reference code. An entry of, say, XN1234 will mean that the user is referred to the kanji with the (unique) Nelson index of 1234. XJ0xxxx and XJ1xxxx are cross-references to the kanji with the JIS hexadecimal code of xxxx. The `0' means the reference is to a JIS X 0208 kanji, and the `1' references a JIS X 0212 kanji.

Zxxxxxx -- a mis-classification code. It means that this kanji is sometimes mis-classified as having the xxxxxx coding. In the case of the SKIP classifications, an extra letter code is used to indicate the type of mis-classification. ZPPn-n-n, ZSPn-n-n and ZBPn-n-n indicate mis-classification according to position, stroke-count and both position and stroke-count. (ZRPn-n-n codes are where Jim Breen & Jack Halpern are having a [hopefully temporary] disagreement over the number of strokes.)

If the final field of a line is not an English field, there is a final space. Each reading and information field is therefore bracketed by space characters (which makes it convenient for searches using programs like "grep".)

As far as possible all entries will have their yomikata and readings attached, even if they are a recognized variant of another kanji. This is to facilitate electronic searches using these fields as keys, and should not be taken as a recommendation to use such obscure kanji.


KANJIDIC is used now to build the "kinfo.dat" file which is used by JDIC and JREADER, and by Stephen Chung's JWP. "kinfo.dat" contains the identical information, but in a compressed form and in a structure suitable for fast indexed access.

KANJIDIC is also used in the XJDIC and MacJDic dictionary programs, and a growing number of other programs such as KDRILL and KDIC.


KANJIDIC was originally compiled, and is maintained by:

Jim Breen
School of Computer Science & Software Engineering
Monash University, Victoria, Australia
If you have suggested changes, send diffs [not complete files] with corrections to him.


KANJIDIC is now rather large, and has information in it which is not much use for people who are not studying and researching Japanese orthography. It is still appropriate to maintain it as a useful freely-available compendium of such information.

For people who only wish to use a subset of the information in KANJIDIC, there is a program "kdfilt.c", also available as kdfilt.exe for MS-DOS, which will strip out unwanted fields. Dan Crevier has also released a program (kanjidicSplit) which does the same for MacJDic users. (For users of the JDIC program, the KANJDFIX.EXE utility also strips out unwanted fields prior to building the KINFO.DAT file.)


(some comments by Jim Breen)

KANJIDIC began as two files: jis1detl.lst and jis2detl.lst, which were later merged into a single file.

The first file was compiled initially from the file "kinfo.dat" supplied by Stephen Chung, who in turn compiled his file from a file prepared by Mike Erickson. I originally added about 1900 "meanings" by James Heisig keyed in by Kevin Moore from the book "Remembering The Kanji". I later added the meanings from Rik Smoody's files, compiled when he was working for Sony in Japan. These appear to have been based on Nelson.

The second file was compiled from a complete JIS2 list with Bushu and stroke counts kindly supplied to me by Jon Crossley, to which I added Nelson numbers, yomikata and meanings extracted from Rik Smoody's file.

Theresa Martin was an early assister with this file, particularly with tracking down and correcting many mistranscribed yomikata (the old zu/dzu, oo/ou, ji/dji, etc. problems).

Jeffrey Friedl did a major overhaul in September-October 1992, in which he added the frequency rankings, Halpern codes, SKIP patterns, updated the grading ("G" fields) to reflect the modern Jouyou lists, corrected radical numbers, corrected stroke counts and readings to fall in line with modern usage.

Magnus Halldorsson corrected some erroneous Halpern numbers, and provided them for a lot of the radicals. He provided the list of Heisig indices, which he originally compiled himself, then verified and expanded using lists from Richard Walters and Antti Karttunen. He also passed on to me the list of Gakken indices compiled by Antti Karttunen.

Lee Collins provided the Unicode mappings (see appendix B)

Iain Sinclair has provided the yomikata, meanings and S&H indices of many of the obscure JIS2 kanji.

Christian Wittern, a Sinologist working at Kyoto University, sent me a monster file prepared by Dr Urs App from Hanazono College. From this I have extracted the Four Corner and Morohashi information. Christian also provided the original Pinyin details, which were later replaced. I am very grateful for these significant contributions.

In March 1994 the Morohashi indices were proof-read and corrected by Christian.

Alfredo Pinochet supplied all the Henshall numbers.

Ingar Holst has provided considerable assistance in regularizing the Bnnn and Cnnn radical classifications to remove some errors that were in the original JIS2 file, and to make it all conform to Nelson's classification.

In mid-1993 I withdrew the SKIP codes from the distributed file as it appeared that their presence violated Jack Halpern's copyright on these codes. Jeffrey Friedl contacted Jack about this, and Jack obtained permission from his publisher for the codes to be included subject to the copyright and usage restrictions stated in this document. In March 1994 the Halpern indices and SKIP codes were checked against an extract from Jack's files, and the "Z" mis-classification codes added, again from his files. Jack has also made a lot of useful comments and suggestions about the content and format of the file. I am most grateful to Jack for his permission and assistance, and also to Jeffrey for making the contact.

In May 1995, a number of updates took place. Jeffrey Friedl established contact with James Heisig, and obtained a further set of his indices. I contacted Mark Spahn (via the "honyaku" mailing list) and he kindly provided most of the missing S&H descriptors, and Jack Halpern released to me the SKIP codes of the kanji not in the New Japanese-English Character Dictionary. For all this material I am most grateful.

In August 1995, I added the O'Neill index numbers. These were compiled by Jenny Nazak, David Rosenfeld and myself. Thanks to Jenny & David for their assistance.

In January and February 1996 the Morohashi numbers were checked thoroughly against two important sources: a file of Unicode-Morohashi data (Uni2Dict) which was prepared by Koichi Yasuoka from the allocation in the JIS X 0221 standard, and the review draft of the proposed revision of the JIS X 0208 standard, which was prepared by the INSTAC Committee, and made available in a text file, thus enabling comparisons. All the mismatches between the three files were examined against the Morohashi text, and extensive corrections made to all three files. I am grateful to Koichi Yasuoka and Masayuki Toyoshima for their considerable assistance in this task.

In March 1996 the Korean readings were added. They were provided by Dr Charles Muller of Toyo Gakuen University (acmuller@gol.com), to whom I am most grateful. Chuck's compilation of Korean readings is extremely thorough and scholarly, and I am pleased to be able to incorporate them.

In April 1996 the readings of all the kanji were compared with those in the JIS X 0208 draft, and a number of corrections and additions made.

In May 1996 I carried out a "unification" of the readings of the KANJIDIC and KANJD212 files, wherein all the readings of the "itaiji" were brought into line. The identification of these itaiji was drawn from a file posted to the fj.kanji group by Taichi Kawabata (kawabata@is.s.u-tokyo.ac.jp), which was compiled at the ETL from the itaiji identification in the JIS X 0208 and JIS X 0212 standards. I corrected a few errors, and added some extra sets which were indicated in the JIS X 0208-1996 draft.

In July 1996 the Pinyin details were completely replaced by a new set. The original Pinyin were from an earlier compilation by Christian Wittern, and and contained many errors. Two more reliable sources had become available: the Uni2Pinyin file compiled by Koichi Yasuoka, which is based in part on the TONEPY.tit by Yongguang Zhang; and the PYCHAR set of readings of Big5 hanzi compiled by Christian Wittern. The Pinyin currently in the KANJIDIC file is a combination of the two, following the order in the Uni2Pinyin file.

In August 1996 I corrected a few more missing and erroneous Nelson numbers, using a massive Nelson list prepared by Wolfgang Cronrath. He also flagged the kokuji, so I added these to the readings fields as "{(kokuji)}".

Also in August 1996 I deleted the handful of former "XJxxxx" cross-references, and replaced them with a much more comprehensive set, so that they now represent all the recognized "itaiji". The file I used for this was the corrected itaiji file mentioned above.

In April 1997 I corrected a large number of bushu codes. Many of these had been identified as errors by Jean-Luc Leger (reiga@iria.mines.u-nancy.fr) who analyzed and examined all the Nelson bushu. I also identified and added a large number of missing Cnnn codes.

Also in April 1997 I added the S&H "Kanji & Kana" indices. These had been keyed by Olivier Galibert (Olivier.Galibert@mines.u-nancy.fr). (There must be an outbreak of kanji interest on Nancy.)

In February 1998, the long-awaited inclusion of the "New Nelson" numbers took place. I had been waiting for the editor of the New Nelson, John Haig, to supply a list (as he had agreed some years before), but in the meantime, Jean-Luc Leger keyed a list, so they are now available.

Also between December 1997 and February 1998 a large number of Level 2 kanji had their stroke counts corrected to bring them into line with the counting principles used in the Level 1 kanji. This usually aligned the counts with those used in the New Nelson and in S&H. Appendix E of this document was amended to reflect this. The leg-work in tracking this material down was done by Wolfgang Cronrath.

During December 1998 & Jan 1999 I updated the stroke counts of many of the Level 2 kanji, using an analysis of them carried out by Wolfgang Cronrath. I also added the De Roo codes, which had been keyed by Jasmin Blanchette, who also typed the explanatory material. I contacted Fr De Roo in Tokyo who readily agreed to the inclusion of thecodes.


In March 2000, James William Breen assigned ownership of the copyright of the dictionary files assembled, coordinated and edited by him to the The Electronic Dictionary Research and Development Group at Monash University.

Information about the formal usage arrangement for EDICT can be found on the Group's WWW page at: http://www.dgs.monash.edu.au/edrdg/

In summary, KANJIDIC can be used, with acknowledgement, for any free software or server, or included in file and software distributions at a nominal charge for the distribution medium. It is also available under non-exclusive licence for commercial usesi, subject to the provisos below.

The following people have granted permission for material for which they hold copyright to be included in the files, and distributed under the above conditions, while retaining their copyright over that material:

Jack HALPERN: The SKIP codes and Frequency codes in the KANJIDIC file.

With regard to the Frequency codes, Mr Halpern stated as follows:

"The commercial utilization of the frequency numbers is prohibited without written permission from Jack Halpern. Use by individuals and small groups for reference and research purposes is permitted, on condition that acknowledgment of the source and this notice are included."
With regard to the SKIP codes, Mr Halpern draws your attention to the statement he has prepared on the matter, which is included at Appendix F.

Christian WITTERN and Koichi YASUOKA: The Pinyin information in the KANJIDIC file.

Urs APP: the Four Corner codes and the Morohashi information in the KANJIDIC file.

Mark SPAHN and Wolfgang HADAMITZKY: the kanji descriptors from their dictionary.

Charles MULLER: the Korean readings.

Joseph DE ROOO: the De Roo codes.


For full information about JIS codes, please see Ken Lunde's "japan.inf" file, or his book "Understanding Japanese Information Processing", O'Reilly 1993. The following is a brief extract from the "japan.inf" file.

"The Japanese character set as described in the document JIS X 0208-1990 specifies 6,879 standard characters; 6,355 kanji in 2 levels (Level 1: 2,965 kanji arranged by pronunciation; Level 2: 3,390 kanji arranged by radical), 86 katakana, 83 hiragana, 10 numerals, 52 Roman characters, 147 symbols, 66 Russian characters, 48 Greek characters, and 32 line elements (for making charts).

This standard was first established in 1978, modified for the first time in 1983 (character position swapping, glyph changes, and four kanji appended to JIS Level 2), and modified again in 1990 (two kanji were appended to JIS Level 2). This character set is widely implemented on a variety of platforms. Encoding methods for JIS X 0208-1990 include Shift-JIS, EUC, and JIS."


The following information about Unicode was provided in 1992 by Lee Collins at Taligent.

(The Unicode sequences are) "the final, official mapping to JIS of the CJK-JRG's (Chinese, Japanese, Korean- Joint Research Group) "Unified Repertoire and Ordering Version 2.0" which is the unified Han character set of ISO 10646 and Unicode. All of the Unicode companies (Apple, IBM, Microsoft, NeXT, Taligent, etc) are now using this mapping. There has been some confusion because of difference in nomenclature. Unicode people call it UniHan, the Chinese sometimes call it HCS (Han Character Set) and ISO calls it "Ideographic CJK Character Unified Repertoire and Ordering". ISO can't use the term "Han" character because Japan was very sensitive to this (even though it is a direct translation of "Kanzi") and it can't be called a character set because only ISO WG2 is empowered with the authority to encode characters. Problems of naming aside, they are all the same thing.

The CJK-JRG was formed under the aegis of ISO in 1990 to investigate and propose a unified Han character set for inclusion in ISO 10646. It brought together various experts on Han characters from China, Hong Kong, Japan, Korea, Taiwan and the United States selected by the national bodies participating in ISO WG2.

Including the initial work in the US on Unicode and in China on GB 13000, which were merged and became the basis for the URO, the task spanned about 4 years. The work was completed in April of this year. It contains 21,000 Han characters from all of the major standards used in East Asia, including JIS X 0208-1990 and JIS X 0212-1990. The Unicode consortium provides a cross-reference file for all of the source sets. To get a copy contact

Steve Greenfield

For further details about the URO/UniHan, you might want to pick up a copy of the "The Unicode Standard Version 1.0 Vol II". It's published by Addison Wesley, ISBN 0-201-60845-6. It's been available in the USA for over a month now. For a slightly different presentation of the characters, a copy of 10646 or of the "Ideographic CJK Character Unified Repertoire and Ordering Version 2.0" might be available through the the Australian national body to ISO WG2."



[This document contains the text and examples from the covers of the "New Japanese-English Character Dictionary" edited by Jack Halpern and published by Kenkyusha and NTC. It is reproduced with Mr Halpern's kind permission.

The text on which this is based used four patterns which are not able to be
reproduced in this document. They are referred to below as #1 through #4,
and relate to the following shapes in the NJECD:

. . . . . . á



A. Determine the SKIP number of your character.


Determine to which of the four PATTERNS your character belongs to get the first part of the SKIP number (the PATTERN NUMBER).

If your character belongs to pattern #1, #2 or #3 (ꢪ#1), carry out the steps in the left column; if it belongs to pattern #4 (#4), carry out the steps in the right column. (REF: R4. How to Identify the Pattern)

. #1 #2 #3 #4

STEP 2 DIVIDE CHARACTER OMIT Divide the character into two parts at (Since solid characters the first division point. [=+] cannot be divided, go to REF: R5. How to Divide the Character STEP 3.) REF: R6. How to Subclassify the Solid Pattern

STEP 3 COUNT STROKES OF SHADED PART DETERMINE TOTAL STROKE-COUNT Count the strokes of the SHADED PART Determine the total stroke-count of to get the second part of the SKIP your character to get the second part number. [ #1 1-4-] of the SKIP number. [ #4 4-3-] REF: Appendix 2. How to Count Strokes REF: Appendix 2. How to Count Strokes

STEP 4 COUNT STROKES OF BLANK PART IDENTIFY SOLID SUBPATTERN Count the strokes of the BLANK PART Determine to which of the four to get the third part of the SKIP SOLID SUBPATTERNS your character number. [ #1 1-4-5] belongs to get the third part of the REF: Appendix 2. How to Count Strokes SKIP number. Select from: `' 1, `' 2, `|' 3, or `' 4. [ #4 4-3-1] REF: R6. How to Subclassify the Solid Pattern

After determining the SKIP number of your character, locate your character entry in one of two ways:

1. Determine the entry number in the Pattern Index beginning on p. 1952 then locate your character entry in the main part of the dictionary. See R3.1.2 Index Method for details.

2. Locate your character entry directly (without referring to the Pattern Index) from its SKIP number. See R3.1.3 Direct Method for details.

NOTE: All references preceded by a section mark (R) refer to SYSTEM OF KANJI INDEXING BY PATTERNS beginning on p. 106a



#1 Characters that can be divided into left and right parts RIGHT: 4-5 Ȭ 1-1 1-11 3-3 WRONG: 1-3 1-4 3-2 ¿ 3-3

#2 Characters that can be divided into top and bottom parts RIGHT: 1-1 3-3 2-3 5-4 WRONG: 1-2 4-2 8-4 4-3

#3 Characters that can be divided by an enclosure element RIGHT: 3-8 3-2 8-3 3-5 WRONG: 1-1 4-3 ̾ 3-3 5-4

#4 Characters that cannot be classified under patterns #1, #2, or #3 RIGHT: 8-1 ʼ 5-2 4-3 Ϳ 3-4 WRONG: 2-1 4-1 4-3




#1 Going from left to right, divide at the first space RIGHT: 4-4 1-2 3-3 WRONG: 2-1 9-3

#2 Going from top to bottom, divide at the first space, horizontal line, or frame element, whichever comes first RIGHT: 1-2 2-8 3-4 2-3 WRONG: 2-1 6-4 2-5 1-2

#3 Going from the outside toward the inside, divide after the first enclosure element RIGHT: 3-6 3-8 8-3 3-2 WRONG: 7-2 11-5

DO NOT VIOLATE THE PRINCIPLE OF ELEMENT INTEGRITY . 1. Never break through strokes . RIGHT: 3-2-2 WRONG: 1-1-4 . 2. Never break through indivisible units . RIGHT: 1-3-8 WRONG: 1-1-10 . 3. Never make unnatural divisions . RIGHT: 3-4-2 WRONG: 2-2-4



`T' 1. Characters that contain a top line RIGHT: 8-1 3-1 6-1 8-1 WRONG: 2-1 3-2 8-1 ʼ 5-1

2. Characters that contain a bottom line RIGHT: 3-2 ʼ 5-2 8-2 WRONG: 3-2 5-2 8-2

3. Characters that contain a through line RIGHT: 4-3 8-3 4-3 WRONG: 4-3 3-3 4-3 7-3

4. Characters that do not contain a top line, bottom line, or through line RIGHT: Ϳ 3-4 3-4 7-4 WRONG: 6-4 3-4 ͧ 4-4 6-4



The Four Corner System has been used for many years in China and Japan for classifying kanji. In China it is losing popularity in favour of Pinyin ordering. Some Japanese dictionaries, such as the Morohashi Daikanwajiten have a Four Corner Index.

The following overview of the system has been condensed from the article "The Four Corner System: an introduction with exercises" by Dr Urs App, which appeared in the Electronic Bodhidharma No 2, February 1992, published by the International Research Institute for Zen Buddhism, Hanazono College. (More examples will be added from that article in due course.)

1. Stroke shapes are divided into ten classes:


2. The Four Digits are derived from the Four Corners in a Z-shaped order.

. A B 7 1 7 7 . for example: . C D 2 9 2 2

Some examples: 2421 2122 7121 2733 0762 Ʊ 7722 4292

3. A shape is only used once. If it fills several corners, it is counted as zero in subsequent corners.

Some examples: 6000 8060 ʬ 8022 2003 2690 6066 0096

4. When the upper or lower half of a character consists of only one (single or composite) shape, it is, regardless of its position, counted as a left corner. The right corner is counted as zero.

Some examples: Ω 0010 ͳ 5060 1017 0022 0024 2090 2050

5. When there is no additional element to the four sides of the characters ., , (and sometimes ), whatever is inside these characters is taken for the lower two corners.

Some examples: 7760 6080 Ԣ 6015 6010 7744 1060 2110

6. The analysis is based on the block-style handwritten kaisho (ܴ) shape of characters.

(This needs attention, as is 3027, not 1027. The top stroke is treated as a Ц.)

7. Some points to note when analyzing shapes:

o Shape 0:

When the horizontal line below a DOT shape (number 3) is connected to another stroke at its right-hand end (as in , etc.) it is not counted as a LID (number 0) but as a DOT.

Examples: 3040 3520 3222

o Shape 6:

Characters such as and where one of the strokes of the square extends beyond it, are not considered to be square (number 6) shapes, but corners (number 7).

Examples: 7710 3222 7710 8377 3010

o Shape 7:

Only the cornered end of corner shapes (number 7) is counted as 7.

Examples: 7171 7222 2762 ȿ 7124

o Shape 8:

Strokes that cross other strokes are not counted as shape number 8 (Ȭ).

Examples: 8043 7743 4003 8043 2143 9043

o Shape 9:

Shapes resembling shape 9, but featuring two strokes in the middle (as in the top part of or ) or two strokes on one side (as in or the bottom part of ) are not considered as 9 shapes.

Examples: 4433 3290 3214

8. Some points to note when choosing corners.

- when a corner is occupied by more than one independent or parallel strokes, the one that extend furthest to the left or right is taken as the corner, regardless of how high or low it is.

examples: 1111 2124 0013 0022 3421 4721

- if there is another shape above (or, at the bottom of the character, below) the leftmost or rightmost stroke of a character, that shape is given preference and is taken as the corner.

examples: 3090 4040 6020 4040 3521 ¶ 4480

- when two composite stroke shapes are interwoven and each could be regarded as a corner, the shape that is higher is taken as the upper corner, and the lower stroke as lower corner.

- when a stroke that slopes downwards to the left or right is supported by another stroke, the latter is taken as the corner.

examples: 2740 ΢ 0073 1962 4464 4410 3424

- a left slanting stroke on the upper left is taken for the left corner only; for the right corner one takes a stroke more to the right.

examples: 2740 ̶ 2350 6752 Ū 2762 2762 2772

9. Shape variations: (Dr App includes several pages of examples)

10. The fifth corner:

In order to differentiate between the several characters with the same code, an optional "fifth corner" is sometimes used. This is, loosely, a shape above the fourth corner which has not been used in any other shape.


These rules apply:

  1. to the stroke-counts themselves;

  2. to the stroke counts in the SKIP codes. Where this results in a SKIP which differs from that in the NJECD, or in the non-NJECD SKIPs provided by Jack Halpern, the Jack Halpern version is included prefixed with "ZR"

The radicals listed below are ones where there are differing approaches to the counting of radicals in the various references. The stroke counting in this file does not strictly follow any reference, but tends to more aligned to Halpern.

  1. B140 KUSA-KANMURI e.g. always counted as 3 strokes (Halpern counts this 4 strokes for the (mostly level 2) kanji where the older form is often printed.) Note that this has been carried through to kanji where this element is not the indexing radical, such as ۯ.

  2. B162 SHIN-NYUU e.g. or counted as 3 or 4 strokes. (Nelson and S&H count it as 2 strokes, and Halpern as either 3 or 4.) [See Note 1 below.]

  3. B163 OOZATOZUKIRI & B170 KOZATO-HEN ˮ and always counted as 3 strokes (Nelson and S&H count it as 2, Halpern as 3.) This also applies where it appears mid-kanji, such as in .

  4. B199 MUGI always counted as 7 strokes, except for & where it is counted as 11. (Nelson and Halpern do the same, and S&H avoid treating it as a radical, but count it as 12 in the remainder.)

  5. B113 SHIMESU e.g. , is counted as 4 strokes in that form, and 5 strokes in its older form, . 18 kanji are in the 4-stroke form and 20 are in the 5-stroke form. (Nelson and S&H count it as 4; Halpern counts it as 4 or 5. [See Note 1.])

  6. B184 SHOKU HEN , , etc.is counted as 8 strokes in the form, and as 9 strokes in the Ҭ and forms. (Nelson and S&H count it as 8 strokes, and Halpern as 8 or 9.) [See Note 1. below.]

  7. B131 SHIN/KERAI . Counted as 7 (Nelson counts it as 6, Halpern as 7 (in the book), and S&H as both for different kanji.)

  8. B136 MAI ASHI . Counted as 7 (traditionally counted as 6, in accordance with the older writing of `'. Nelson counts as 6, S&H as 7, and Halpern as 7 for and ̾Ѵ and 6 for the rest.) Note this is also applied to counting and for kanji with the pattern.

  9. B131 SHIN or KERAI . Counted as 7 (traditionally counted as 6). Nelson counts as 6, Halpern as 7, and S&H as 6 or 7 in different cases.

  10. The ROO or OI radical (Ϸ) has a variant consisting of the top 4 strokes. For example, it is in . Traditionally, this variant had an extra dot, and was counted as 5 strokes. I'm counting it as 4 throughout.
  1. While the pattern is a 6-stroke radical, the top half of is made up of three distinct parts totalling 8 strokes. Note that this also is the case with տ, , and despite the simplification in the JIS glyphs.

  2. (KIBA HEN) is a problem. It is classically counted as 4 strokes, but these days has a flick that makes it effectively 5. Halpern, Nelson and S&H usually have it as 5 strokes, so I'm standardizing on that.

  3. Another little horror is (MU or NASHI), which is classically counted as 4 strokes. The most common variant has 5 strokes, but looks like 6. Halpern, S&H and the Classical Nelson count this as 4 strokes, and the New Nelson as 5. I'm making it 5 too.

  4. The JUU or ASHIATO radical is at the bottom of and . It is traditionally counted as 5 strokes, although sometimes it looks like 4. I'm using 5 throughout.

  5. A related shape is , as in , , , etc. This is sometimes counted as two strokes (both Nelsons) and sometimes as three strokes (Halpern, S&H). Classically it is regarded as two strokes. I am using 6 strokes for .

  6. The pattern to the left of , which appears in several kanji, e.g. ʾ and , has 8 strokes. (There are 3 strokes at the top as in .)

  7. The "east" pattern () has 8 strokes. There is an older form in which there are two strokes in the box (). It is counted as 8 strokes here in the form (e.g. ) and 9 in the form, as in .

  8. The pattern at the bottom of is counted as 4 strokes in modern dictionaries, although traditionally it was 5.

  9. The pattern , which appears in several kanji, is counted as 9 strokes. Several dictionaries count it as either 8 or 9.
Note The JIS X 0208-1990 standard does not formally specify the precise glyphs used for kanji, however the glyphs it uses in the published version have become de facto standards for many font compilations. In the published standard, for several kanji, e.g. é/, /, /Ҭ, the JIS level one kanji use the simpler form, and the Level 2 kanji use the older more complex form. Just to make matters worse, many fonts for JIS X 0208 kanji are based on the bit-maps specified in JIS X 9051-1984 standard, which defines the 16x16 patterns for JIS X 0208-1983 characters. According to Ken Lunde: "This standard was not very good, and JSA is no longer supporting it." Anyway, JIS X 9051-1984 had the simpler form for all these bushu in both Levels 1 and 2, as well as having simplifications of kanji like . Thus, as the font foundries have freedom to choose whichever glyphs they like, what you see on your screen may well not agree with these rules. All the rules in this appendix relate to the glyphs as published in the JIS X 0208-1990 standard, and as appearing in font compilations based on them.

APPENDIX F.CONDITIONS FOR USING SKIP DATA by Jack Halpern (jack@kanji.org)

Ever since my New Japanese-English Character Dictionary (NJECD) came out (Kenkyusha 1990, NTC 1993), I have been getting inquiries asking for permission to use SKIP (System of Kanji Indexing by Patterns) data in software products and electronic dictionaries. Below I explain the policy of the Kanji Dictionary Publishing Society (KDPS) on how to use copyright issues when distributing SKIP data or using it in software product or electronic dictionary.


Briefly, SKIP is an indexing system that enables the user to locate kanji quickly and accurately. The system is extremely convenient because it can be learned in a very short time, is easy to use, and requires very little prior knowledge of kanji.

The central idea of SKIP is the classification of characters into four major categories on the basis of easy-to-identify geometrical <patterns>:

1. Left-right
2. Up-down
3. Enclosure
4. Solid
Characters belonging to the first three categories are arranged in ascending order of hyphenated numerals that represent the number of strokes in the <shaded part,> and the number of strokes in the <blank part.> See http://www.kanji.org and NJECD front matter for details.

To distribute SKIP data within a group or use it in a commercial or non-commercial product, please confirm that you agree to the following conditions:


    SKIP data is protected by copyright, copyleft and patent laws. The copyright holder is Jack Halpern, chief editor of KDPS (the Kanji Dictionary Publishing Society). The SKIP data must be protected from illegal copying and distribution, using such meaasures encryption. The data must be encrypted if it is to be used in any kind of product, including commercial products, software and freeware. The data, or extracts from it, must not be distributed to a third party, must not be sold as part of any commercial software package, and must not be incorporated in any published dictionary or other printed document without the specific permission of the copyright holder.


    The source of SKIP data shall be acknowledged in the information screens of the product, and the following disclaimer should appear in the documentation and/or help screens:

    "SKIP (System of Kanji Indexing by Patterns) numbers are derived from the New Japanese-English Character Dictionary (Kenkyusha 1990, NTC 1993) and The Kodansha Kanji Learner's Dictionary (Kodansha International, 1999). SKIP is protected by copyright, copyleft and patent laws. The commercial or non-commercial utilization of SKIP in any form is strictly forbidden without the written permission of Jack Halpern, the copyright holder. Such permission is normally granted. Please contact jack@kanji.org and/or see http://www.kanji.org."


    SKIP is a product of seven years of computer-assisted research and experimentation on how kanji elements are intuitively perceived in terms of their parts. Development work was financed by private funds and research grants. To enable us to continue to develop useful data and products, we ask for you cooperation by paying KDPS (the Kanji Dictionary Publishing Society) a royalty 0.5% (negotiable) if you are using the data for a commercial product. Depending on the circumstances, it is also possible to use SKIP data free of charge or at a lower royalty.

    Finally, please send a copy of your product to Jack Halpern



[This document contains the text found in the second edition of "2001 Kanji" edited by Joseph R. De Roo and published by Bonjinsha.]

The system used in "2001 Kanji" is intended for the beginner who encounters a kanji and wants to look it up, knowing neither its radical, pronunciation, nor its exact number of strokes. The method consists of looking at the top of the kanji, and then at its bottom, disregarding its other parts.

"2001 Kanji" provides drawings for all graphic elements. This information cannot be reproduced here. However, an attempt was made to describe each element as much as possible given the constraints of a computer text file, and examples of characters possessing the element are always given.

Two-step visual method for locating a kanji:

1. Observe its EXTREME TOP or LEFT TOP.

There are only four possibilities: DOT (Ц), VERTICAL LINE (), DIAGONAL LINE (), HORIZONTAL LINE (). Each of these four strokes can occur either in isolation or in connection with one or more strokes. Each of the four groups of graphic elements correspond to the four basic strokes in their immediate environment. Each element has a number wich will become the first half of the kanji number.

DOT (Ц): 3 DOT (Ц) ɬ 4 ROOF (е) 5 DOTTED CLIFF () ģ 6 ALTAR Ƿ 7 KANA U () 8 LID 9 HORNS

VERTICAL LINE (): 10 SMALL ON BOX Ⱦ 11 SMALL () ɹ ϧ 12 VERTICAL LINE () Ҹ 13 HAND TO THE LEFT 14 CROSS () ¸ ˮ 15 CROSS ON BOX () ī « 16 KANA KA () 17 WOMAN () 18 TREE () 19 LETTER H (װ) ʦ ¶

DIAGONAL LINE (): 20 KANA NO () ˳ ë Ȭ 21 MAN TO THE LEFT () 22 THOUSAND () 23 MAN TO THE TOP ̵ 24 COW () 25 KANA KU () ұ 26 HILL TOP α 27 LEFT ARROW () 28 ROOF () 29 X ()

HORIZONTAL LINE (): 30 HORIZONTAL LINE () Ʀ 31 FOURTH () ŷ ʿ 32 BALD (Ѻ) 33 CLIFF () ä ȿ 34 TOP-LEFT CORNER Ĺ 35 TOP-RIGHT CORNER ȯ ͽ λ ׮ 36 UPSIDE-DOWN CAN () Ʊ ð 37 MOUTH () ­ ̱ 38 SUN () ¨ 39 EYE TOP


There are nine possibilities: DOT (Ц), LEFT HOOK (Э), VERTICAL LINE (), RIGHT HOOK, DIAGONAL LINE (), BACK DIAGONAL LINE (), BOTTOM OF HEAD ɥ, BOTTOM OF WATAKUSHI , HORIZONTAL LINE (). They are listed in association with one or more strokes. The number of the bottom element will become the second half of the kanji number.

DOT (Ц): 40 FOUR DOTS ̵ 41 SMALL () ; 42 WATER () ɹ

LEFT HOOK (Э): 43 KANA RI () 44 SEAL () ʦ Ĥ 45 SWORD BOTTOM () ǵ 46 MOON () ͭ 47 DOTLESS INCH в ͷ Ϳ ð 48 INCH () 49 MOUTH LEFT HOOK 50 BIRD BOTTOM Ļ 51 ANIMAL () ʪ 52 BOW BOTTOM ұ 53 LEFT HOOK (Э) λ




BACK DIAGONAL LINE (): 61 SMALL PODIUM ϻ 62 BACK KANA NO () 63 BIG () ŷ 64 TREE () « ̤ 65 SMALL SPOON ι Ĺ ä 66 GOVERN (Щ) ʸ ھ 67 AGAIN () ׮ 68 WINDY AGAIN () 69 WOMAN ()




The number of the kanji you are looking for consists of the top number coming first and the bottom number coming second, the two numbers being placed side by side. E.g., 363 (3 63), 747 (7 47).

There are two rules always to keep in mind:

a. Ignore the complete enclosure and the "road" radical (as in ƻ). Look at the top and bottom (in some cases only the bottom) of what is inside the complete enclosure, and of what is to the upper right of "road". E.g., 1262, 2177, ƻ 979, ˥ 2755.

b. When a part is enclosed by the "gate" radical, take the bottom or right bottom of that part. E.g., Ʈ 3848, 1864.