This paper uses clickstream data from Plurimus Corp. (formerly Foveon Corp.) to analyze user choice of Internet portals. It will show that commonly used econometric models for examining grocery scanner data can be applied to clickstream data on advertising-based online markets. Developing a framework to study consumer choices of free websites is an essential step to better understanding user behavior on the Internet. The main data for this study is a clickstream data set consisting of every website visited by 2654 users from December 27 1999 to March 31 2000. Using this data, I construct several variables including search success, time spent searching, and whether a website is an individual's starting page. I also have advertising and media mentions data. This study helps to increase understanding of user behavior on the Internet. It explores some key determinants of website choice and simulates market responses to changes in the online environment. By better understanding website choice, we can better evaluate the implications of policy decisions.