Commonly used measures of effect, such as risk ratios and odds ratios, may be quite biased when used to assess the effect of factors that alter transmission risks given exposure to infected individuals. This is demonstrated in a simulation model involving a higher-risk behavior and a lower-risk behavior affecting the sexual transmission of human immunodeficiency virus. The bias arises because population contact patterns between higher-risk and lower-risk persons change their relative probabilities of exposure to an infected individual as an epidemic progresses. The assessment of contact patterns is thus central to risk assessment for contagious diseases. A new formulation of selective mixing presented here, together with a structured mixing specification of the social settings of contact, provides a theoretic framework for the investigation of contact pattern determinants.