To identify biases and threats to validity of Internet survey data collection on HIV-related risk behaviors, we studied 1,546 Latino men who have sex with men on the Internet recruited through banner impressions on a leading national gay Internet site. The study could be completed in English or Spanish. Of those commencing, 33.6% dropped out before completing the 450-field questionnaire. None of the linguistic variables (level of use of Spanish or English) predicted dropout. However, dropouts were more likely to identify as Puerto Rican or Black, to reject the $20 compensation or offer it to a charity, to not have met men for sex on the Internet, to identify as bisexual or heterosexual, and to use Web sites or personal ads for contact and to use the Internet less at home than those who completed the study. Men in seroconcordant monogamous relationships and those who had not met a man for sex on the Internet were also more likely to drop out. These data suggest that there are no linguistic and few demographic and Internet use variables that are associated with dropout. Issues of compensation and respondent characteristics that make it likely that there will be a large number of inapplicable data fields in the questionnaire appear to be significant predictors of dropout. Although there were many data missing, the dropouts did not appear to be at greater HIV-associated risk than the completers. The fact that there appear to be few systematic demographic or Internet use biases in dropouts suggests that the completers do not represent a seriously skewed sample of those Latinos who commence the Internet survey.