Taking advantage of online customer reviews for recommendation system is becoming increasingly important in e-commerce field due to rich implication information of reviews. By analyzing the sentiments and topics through these reviews, a set of sentimental features (SF) can be exploited to represent customer preference. In this work, we firstly construct user-SF matrix instead of traditional user-item matrix, and use such matrix to derive the nearest neighbor users whose preference is in accordance with target user. Then a list of recommended items are ranked in order to select top-N items for recommendation. Based on the proposed framework, we believe that this approach can enhance the performance of recommender system with the help of sentimental features.