A series of experiments have been carried comparing the 2002–2006 General Social Survey with data collected from the Web-enabled Knowledge Networks Panel surveys. First, the results indicate that the level of don’t knows are highly contingent on format and layout. It is possible however to design in-person and web surveys to produce similar and comparable levels of item on-response. Second, the substantive distributions are not statistically different across modes for the majority of items. Third, statistically significant and substantively large mode effects do appear for an appreciable minority of items. These differences probably relate to the different demand characteristics.