The article reports on results of a split-ballot experiment conducted in 1987 to test alternative versions of the U.S. decennial census long form. The study was conducted as part of a program of research to improve the design of the long-form questionnaire. Typographic and other layout changes were made to increase the consistency of the form's appearance and clearer instructions were provided on how to complete the census form. In the 1986 form, detailed Hispanic and Asian categories were listed separately. In the revised form, categories were combined, with an instruction to write in each person's specific group. The revised and 1986 forms were compared in a series of split-panel experiments. A total 515 people filled out forms for themselves and members of their households, providing data on 1,446 persons. Results show that despite revisions to the race question and response categories, race distributions are very similar for the two forms. Each form identified about the same percentage of individuals in the categories of white, black, Asian or Pacific Islander, and other race. Most Hispanic immigrants simply do not apply the categories "white" or "black" to themselves, regardless of whether or not they have first reported their Hispanic origin.