This article describes a novel and fast recommender system for websites based on product taxonomy and user click patterns. The proposed system consists of the following four steps. First, a product-preference matrix for each customer is estimated through a linear combination of the click, basket placement, and purchase statuses. Second, the preference matrix of the genre and that of the specific type are constructed by computing the ratios of the number of clicks, basket placements, and purchases of a product to the totals. Third, cluster analysis is performed using the genre preference matrix, and a neighborhood formation process is conducted using the specific-type preference matrix. Finally, data are generated for prediction, in which a customer's preference for specific types is greater than a given threshold value. Using these data sets, computational burden and processing time are greatly reduced. The effectiveness of the proposed approach was assessed by applying the F1 metric to an experimental e-commerce website. The proposed method outperformed conventional methods.