The invention relates to a similar articles recommendation method based on a theme model. Firstly, an original text preprocessing is conducted, a simple content of an article is extracted; then, the content of the article is divided into words, an analysis of the properties of the words is conducted, noun phrases are filtered out, a word bag is extracted, a main word feature vector is formed; next, a TFIDF model is trained with the word feature vectors of all articles, based on the TFIDF model, the word feature vector of each article is calculated, a TFIDF feature vector is formed; again, an LSI theme model is trained with the TFIDF feature vectors of all articles; finally, the LSI model is used to calculate, potential theme feature vectors of the article are obtained, similar articles canbe obtained from the calculation of the vector similarity. The method can help Internet users to find interesting articles effectively, and has the advantages of a wider range of application, lower manual marking cost, better recommendation diversity and the like.