Collaborative filtering (CF) technique is the most successful method for recommendation system. In this article, we developed a fashion recommendation system by using CF technique. In order to improve on data sparseness problems in CF technique, firstly we built users’ similarities based on users’ background information which is related with fashion, then the neighbors’ predicting ratings were filled into the U-I rating matrix in advance before the traditional collaborative filtering. While computing the background information similarities, we develop a hybrid similarity model which can deal with different types of properties. The method can solve the data sparseness of U-I rating matrix effectively.