In this paper, we present the development of a personalized Internet shopping agent, Oh!Hot, which can provide intelligent assistance according to the online behaviors of users, such as registration, browsing, querying, purchasing, etc. Oh!Hot employs a hierarchy of ”User Preference Matrices (UPM's)” technique to provide efficient and accurate personalized services. Each UPM records the online behaviors of users on a specified scope of merchandises to reflect the up-to-date degrees of user interest, from which the merchandise information can be recommended to the most potential users. Based on the experimental results from practical applications, it can be seen that Oh!Hot is capable of finding the most desirable merchandise information for users as well as predicting user interests on merchandises.