The news topic discovery model based on multi feature fusion text clustering was proposed fusing multi features of news, such as named entities, news headlines, important paragraphs, text semantics and so on. Based on the multi feature influence of news, a multi feature fusion text clustering method was put forward in this model. In this way, vector space model and similarity algorithm based on feature words, news headlines, important paragraphs were constructed, subject space model and similarity algorithm based on latent Dirichlet allocation were constructed, named entity model and similarity algorithm based on named entities were constructed, and those three similarity algorithms were fused optimally. Based on multi feature fusion text clustering method, the Single-Pass algorithm used in the news topic discovery was improved. Experiments were carried out on the real news data set, and the experimental results show that the model can improve the accuracy rate, recall rate and comprehensive evaluation index of the news topic discovery, and have some ability of self-adaption.