Sample coverage has dominated the discussions of non-observation error in web surveys to date, and is a key determinant in the selection of mode. As web surveys improve and coverage increases, our attention shifts to nonresponse. In the Spring of 2001, the Student Life Survey was conducted among 7000 randomly selected undergraduate students at the University of Michigan. The UM Student Life survey is conducted every other year and was designed to measure attitudes and behaviors regarding drugs, alcohol and other student topics. The sample was randomly split into two equal survey mode groups: web and mail. A small set of demographic variables for the entire sample was provided by the University, and allowed us to evaluate and compare how nonresponse patterns differed between modes. After the data was collected via web and mail, a rigorous follow up of a sample of nonrespondents from each mode group was conducted by telephone to ascertain their reason(s) for nonresponse. We will discuss the reported reasons for nonresponse. We will discuss how these results could be used to evaluate the best possible mixed mode design for this population, in an overall effort to reduce costs and error.