This paper examines how beliefs about own HIV status affect decisions to engage in risky sexual behavior, as measured by having extramarital sex and/or multiple sex partners. The empirical analysis is based on a panel survey of males from the 2006 and 2008 rounds of the Malawi Discussion and Ideational Change Project (MDICP). The paper first develops a behavioral model of the belief-risky behavior relationship. It then estimates the causal effect of beliefs on risky behavior in a way that takes into account the belief updating mechanisms implied by the model. In particular, the Arellano and Carrasco (2003) semiparametric panel data estimator that is used accommodates both unobserved heterogeneity and belief endogeneity, arising from dependence of current beliefs on past risky behavior. Results show that downward revisions in the belief assigned to being HIV positive increase risky behavior and upward revisions decrease it. We estimate for example that a change in the perceived probability of being HIV positive from 0 to 100% reduces risky behavior between 13.7 and 36.4 percentage points depending on the risky behavior definition and year. Implementation of a modifed estimator that allows for misreporting of risky behavior finds the estimates to be downward biased but relatively robust to a wide range of plausible misreporting levels.