Abstract: Collaborative filtering algorithm is one of the most important technologies in electronic commerce recommendation system．The accuracy of recommendation system directly depends on the effectiveness of the similarity measure．The methods of traditional similarity measure mainly focus on the similarity of user common rating items,but ignore the relationship between the user common rating items and all items the user rates．The relationship between the user common rating items and all items the user rates can be calculated by Tanimoto coefficient．However,Tanimoto coefficient is based on the mode of binary operation,which will not get the satisfactory result if it is directly applied in recommendation system．Aiming at the above problems,the improved Tanimoto coefficient was proposed,and the relationship between the user common rating items and all items the user rates was blended into the traditional similarity measure methods．Experiments show that,to a certain extent,the proposed collaborative filtering algorithm is more effective and accurate.