Aiming at such the problems of sparse data and non-currency to select the nearest neighbors, a collaborative filtering recommendation algorithm of sub-similarity integration between items is proposed in the paper. According to every attribute value of the target user, the users whose attribute value is the same as target user's are selected as user's sub-space, similarity(sub-similarity of items) between the target item and others in the user's sum-space is calculated. Based on it, according to sub-similarity of items, k-nearest-neighbors are selected to calculate it's prediction value. Finally, weighted sum of prediction value of user's attributes is calculated to get final prediction value of the target item. Experimental result shows that the algorithm can select nearest neighbors of target item correctly and improve recommendation quality of spare data.