In World Wide Web environments, recommender systems are useful to reduce information overloading. A content-based recommender system recommends items according to their features. Vector Space Model (VSM) is a popular way to recommend items that are similar to those the user liked in the past. The main disadvantages of this content-based method are overspecialization and new user problems that incurred by incomplete information on user preferences. Therefore, to construct users' complete preference profiles may enhance the effectiveness of recommender systems. Some utility function elicitation methods have been developed based on Multi-Attribute Utility Theory. Whether these utility-based methods are able to outperform the traditional VSM method for recommendations is investigated in this research. This research adopts the RBFN and SMARTER methods to construct users' multi-attribute utility functions that represent their complete preferences. A laboratory experiment is conducted to compare the utility-based methods with the traditional VSM method in terms of recommendation accuracy, time expense, and user perceptions. The research results demonstrate that the VSM method is suitable to recommend items with mostly nominal attributes, and the SMARTER method is suitable to recommend items with mostly numerical attributes. The RBFN method has reliable accuracy and time expense in both recommendation contexts.