The invention discloses a one-dimensional bar code word decoding method based on the least square method, and the method comprises the following specific steps: calculating the number ei of modules in a similar edge distance of each code word of a bar code according to an actually scanned bar vacancy width, comparing the number ei of modules with the number Ei of module widths in an accurate code word in a standard edge difference table, judging whether the number ei of modules is equal to the number Ei of module widths or not, and calculating the number N when ei is equal to Ei, wherein g is the number of modules in the similar edge distance of the code word; determining that the decoding of the code word is not successful if N is less than g-2 or N is less than or equal to one; searching to-be-selected possible code words Fi if N is greater than or equalt o g-2 or N is greater than 2; selecting the code word with the minimum error Err as the correct code word from all possible code words through the least square method, and carrying out the detection of an error code for the code words; determining that the decoding is successful if the error code detection is passed, or else determining that the decoding is not successful; selecting the possible characters if N is equal to g, and carrying out the related calculation whether the characters are error codes or not. The method can greatly improve the decoding success rate of bar code words.