Movie recommendation systems aim to recommend movies that users may be interested in. In this paper, we introduce a content-based movie recommendation system which can use different feature sets, namely, actor features, director features, genre features and keyword features. For each user, we assign a weight to each feature in a feature set based on the particular user's past behavior. We produce user's implicit rating for a movie based on the duration of the movie that the user viewed. In order to predict a rating for user and a movie, using a particular feature set, we merge the user specific weights of movie's features. We also produce ratings using all feature sets. We evaluate each recommendation method based on precision, recall and F measure on ten movie recommendations.