Data bearing on the effect of language of the interview on Hispanics' responses to a health survey in which no back-translation was undertaken reveal both lower reliabilities and lower bivariate correlations among Hispanics interviewed in Spanish than among Hispanics interviewed in English. An independent back-translation aimed at creating an English version of the questionnaire that was linguistically equivalent to the Spanish version indicated several instances in which the Spanish version was unidiomatic. Differences between the Spanish and English version in the idiomatic quality of the interview items, while not affecting meaning, appear to have affected the seriousness with which the interview situation was perceived. These perceptions, in turn, appear to have led to the response discrepancies observed.