We use an implicit association test (IAT) to measure implicit gender attitudes and examine the malleability of these attitudes using a randomized field experiment and quasi-experimental data from Tunisia. Women that appear most conservative respond to a randomized video treatment by reducing their implicit gender bias. Also, female interviewers invite more conservative responses to the IAT, especially among the male subsample. Perceived religiosity of the interviewer affects self-reported gender attitudes, but not IAT measures, suggesting social desirability may be at work. We discuss the implications of our findings for the use of implicit measures in development research.