《尚書》有伏生一系所傳《今文尚書》，有孔安國所傳壁中《古文尚書》，有晚出隸古定《尚書》，因爲《今文尚書》與壁中《古文尚書》均已失傳，晚出隸古定《尚書》經唐朝衛包改字，所存最早全文爲《唐石經》。故清代學者考定《今文尚書》與壁中《古文尚書》之文字，皆以《唐石經》的《古文尚書》中被認爲是真書的33篇作爲基礎材料。但《唐石經》已被衛包改爲今字，而且距隸古定《尚書》面世已達五百多年，其間輾轉傳抄，應有不少有意改動與無意致誤者。文章通過七例敦煌與日本古寫本的異文，對前賢所論漢代今古文《尚書》文字的結論及其文字的判定標準提出疑問，認爲在判定漢代今古文《尚書》文字時應該充分利用早期寫本文獻，並從文本演變的角度思考《尚書》文字判定的標準問題。

There are 3 versions of Shangshu (尚書), including Jinwen Shangshu (今文尚書) inherited from Fu Sheng (伏生), Guwen Shangshu (古文尚書) excavated in the wall of Confucius' old house and read by Kong An-guo, and the ＂liguding (隸古定)＂edition which is also known as Faked Guwen Shangshu (僞古文尚書). While the first two didn't come down, the last one was withal transformed to a kaishu (楷書) version by Wei Bao (衛包) in Tang Dynasty, which was written in Tangshijing (唐石經), the earliest intact edition of Shangshu that we can watch today. The scholars of Qing Dynasty researched the characters of Jinwen Shangshu and Guwen Shangshu by the Tangshijing version, 33 chapters of which are identified as credible Shangshu texts. However, the characters of this version were turned from Guzi (古字) to Jinzi (今字) by Wei Bao, with a time lag for more than 5 centuries from the appearance of Faked Guwen Shangshu, when intended or unmeant change must take place by different kinds of transcription. This paper researchs different characters in the ancient manuscripts from Dunhuang and Japan with 7 examples, thus challenging the conclusions of former scholars about the characters of Jinwen Shangshu and Guwen Shangshu besides oppugning their method. We claim that the early ancient manuscripts should be abundantly used to judge the characters of different versions of Shangshu, and take the evolution of texts in consideration.