Historically, many researchers have followed the “rule of thumb” that translated survey instruments should be cognitively tested with only monolingual and limited source-language proficient respondents. There are various rationales for this practice, such as assumptions that 1) monolingual respondents are the intended users of translated survey instruments (Pan, et al., 2007); and 2) fully bilingual respondents might be more likely to understand a poorly translated phrase. Bilinguals are also more likely to understand inappropriately literal translations for concepts that do not exist in the target language. Whether pretesting translations with monolingual or bilingual respondents produces different results has rarely been studied empirically and most studies have not been designed to look at this issue specifically (Park, et al., 2014). This paper focuses on results from two different Spanish-language cognitive and usability studies with monolingual and bilingual Spanish-speaking respondents: 1) testing of a U.S. Census Test internet instrument and 2) testing of a U.S. Census Test CAPI instrument. In the first study, a small number of bilingual respondents were inadvertently included in the study. The second study was deliberately designed to compare testing with monolingual and bilingual respondents, with 20 interviews conducted with each type of respondent. Preliminary results from the first study indicate that bilingual respondents uncovered an equal or higher number of translation and usability problems as did the monolingual respondents. One limitation of these results is that the study was not specifically designed to examine this issue and thus bilingual respondents were not systematically recruited. In addition, there were correlations between English-language proficiency, computer proficiency and educational level, with the bilingual respondents having higher levels of each. We will compare the number and types of findings uncovered by monolingual and bilingual respondents in the internet and CAPI studies and will explore mode differences noted across the two studies.