Using a multi-stage approach, the effects of design and question wording on the 1990 census decennial long form on responses were examined. Three methods were used: 1) one-on-one observation and “think-aloud” sessions; 2) experimental group sessions (n = 515) using surveys; and 3) mailout/mailback surveys (n = 5875). With the latter two methods, form design and wording and sequencing of questions varied across versions of the form. Observational session results showed respondents had difficulty navigating the form, as well differentiating between the instructions and the form itself. Experimental group session results showed that the revised form produced a higher rate of understanding of residence rules than the control form, question 1 was often left blank, and location of the coverage question affected item nonresponse. Mailout/mailback survey results showed that the layout of revised forms increased item response rates, item nonresponse rates varied by age, education and race/ethnicity, incorrect data was often entered for the “usual home elsewhere” question, and very few people had trouble with the coverage question.