Item-based collaborative filtering recommendation system has been widely used in many fields, which generates recommendations based on similarity between items. However, the conventional similarity calculation may produce inaccurate results because of data sparsity. To alleviate this problem, this paper proposes a new method of similarity calculation based on item rating and genre. Firstly, similarity calculation based on item rating are proposed, which reduces similarity between items with fewer co-rating users. Genre information is an inherent attribute of an item which could not be changed by user behavior. It reflects the common characteristics among items, then item similarity based on the item’s dependency on genre are calculated. Finally, a trade-off between rating and genre similarity are proposed to calaulate the similarity between items. Experimental results show that the proposed method can alleviate the issue of inaccurate similarity caused by sparse data and improve the recommendation quality.