Vignettes are a common tool to assess how people think about specific concepts, allowing data to be gathered on survey questions that might not normally apply to the respondent (Martin, 2004). They are also useful as “a less threatening way to explore sensitive subjects” (Finch, 1987). To test potential new screening items for the National Crime Victimization Survey (NCVS), Westat presented seven vignettes about incidents of personal victimization to a non-probability web panel. Respondents were asked how they would answer a NCVS screening question based on each vignette. Each vignette had four to six versions in a factorial design, with vignette versions independently randomized for each respondent. Factors varied the seriousness of the crime, who committed the crime, and other conditions. Respondents were assigned to report how  they would respond to current NCVS item wording, vs. a “streamlined” version which included fewer examples of the victimization type. Surprising results were found, with high levels of respondents indicating that they would not consider the vignettes to be a crime, even using the current NCVS wording.To understand why reporting results were lower than expected, we conducted a series of 18 follow-up cognitive interviews. Our hypotheses were 1) that the number of examples and cues in the NCVS items might be leading to cognitive overload; and 2) that respondents may be focusing on a salient characteristic of the vignette which may be mitigating the interpretation of it as a victimization. Cognitive interviews revealed few problems with either of the versions of the NCVS questions. Rather, participants revealed it was the vignettes themselves that caused difficulty. This paper will discuss findings from the survey and cognitive interviews, and will present practical considerations for using vignettes in survey research.