The CATI interviews conducted among ACS-National sample mail nonresponse cases in October and November of 1997 included an experimental test of two different instrument structures — the traditional “person-based” approach versus a new “topic-based” design. A person-based interview in essence completes an entire interview for the first eligible household member, then returns to the beginning and completes an interview for the second person, and so on through all eligible persons. In contrast, a topic-based design gathers data on one “topic” for every person and then proceeds to the next topic, in effect making only one “pass” through the instrument. This paper presents the results of that experiment, comparing the performance of the two instrument designs on multiple dimensions: response/refusal rates, length of interview, assessments of interviewers and their supervisors, respondent evaluations, behavior coding of interviewer/respondent interactions, item nonresponse, and data outcomes including response distributions and within-household response consistency. In most respects, we find the topic-based design clearly superior to the person-based design.