Recent studies have indicated variation in the infectivity β of HIV among heterosexual couples. We represent this heterogeneity by modelling β as a random variable. Using data on the number of contacts and seroconversion of couples, we fit the model by maximum-likelihood estimation with a beta distribution and a discrete distribution for β. The estimates indicate the heterogeneity is extreme. The limitations of the model and the implications of the results for epidemiologic spread and further research are discussed.