The growth of the Internet for shopping has led to an increasing interest in tools for assisting consumers with decision-making, efficiently using the vast quantity of widely dispersed information. Online product recommendation agents gather information from consumers and then match these consumer preferences with their database of products to recommend the best product. Two approaches can be taken for gathering information from consumers on their preferences – conjoint-type full-profile ratings or self-explicated ratings. That is, organizations may infer consumers' preferences for attributes and levels on the basis of their ratings of several alternative products or may simply directly ask them their evaluations of various attributes and levels. We compare these two approaches and find that, in general, they do not result in the same conclusions. In this paper we examine the differences in the approaches to making recommendations and discuss the implications of these differences. Our results show that there is a closer match between the methods for products closer to the extremes of consumer preference. Also, our study shows that a recommendation agent should offer more than one recommendation in order to match the needs of the system user.