The invention provides a multi-language identification method based on coding and machine learning, and belongs to processing technology of computers to natural languages. The method comprises the following steps: performing language identification on a text through a machine learning unit and a coding identification unit, counting the vocabulary of each language during the coding identification, outputting a single identified language when the identification result of the machine learning unit is within a judgment interval of the coding identification unit and when the languages identified by the two units are consistent, when the coding identification unit identifies a plurality of languages, judging a mixed language rule, and if the vocabulary of a second language in the text reaches a set proportion, judging the text as a mixed language. According to the multi-language identification method provided by the invention, a long text can be judged after random sampling to improve the identification efficiency. By adoption of the multi-language identification method provided by the invention, 99 languages can be identified accurately and efficiently, such as Chinese familiar style and complex font, Japanese, French, English and the like, and the mixed language text recognition is supported, so the multi-language identification method has a wide application prospect in mass data analysis and public opinion monitoring.