This study utilizes a processing tracing method to explore the processes of trust formation in web-based productbrokering recommendation agents (RAs). We compare and contrast the processes of trust/distrust formation in an attribute-based RA (a typical content-based RA) versus a need-based RA (a content-based RA plus need-based questions). Concurrent verbal protocols from 49 subjects were collected, transcribed, and analyzed. Our protocol analysis results show that the need-based RA elicits significantly more trust formation processes and fewer distrust formation processes than the attribute-based RA does, which explains why the level of customer trust in the need-based RA is significantly higher than the level of customer trust in the attribute-based RA. Interestingly, our results show that, for both types of RAs, the top three processes of trust formations are different from the top three processes of distrust formations. Suggestions are given on how to design more trustworthy RAs.