The objective of this study was to investigate the relationship between the number of unprotected heterosexual contacts with an HIV-infected person and the probability of HIV transmission. Data from a European study involving 563 heterosexual partners of HIV-infected subjects were analyzed. The number of unprotected contacts could be estimated for 525 couples (377 with male index case, 148 with female index case) from the reported frequency of unprotected contacts and an estimate of the length of the period during which transmission could have occurred. Nonparametric (isotonic regression) and parametric (Bernoulli model) analyses were performed on data at study entry and on follow-up data (121 couples). The nonparametric analysis resulted in several exposure groups, with the proportion of infected partners increasing with the number of contacts. For example, the percentage of female partners infected ranged from 10%, among those with < 10 unprotected contacts with an infected male, to 23% after 2,000 unprotected contacts. The parametric estimates of (assumed constant) per-contact infectivity were higher for male-to-female than for female-to-male transmission, but not significantly so. However, in comparison with nonparametric estimates, the model assuming constant infectivity appears to seriously underestimate the risk after very few contacts and to seriously overestimate the risk associated with a large number of contacts. Our results suggest that the association between the number of unprotected sexual contacts and the probability of infection is weak and highly inconsistent with constant per-contact infectivity. Probable explanations for these findings include large variability in infectivity between couples and within individuals over time. Estimates based on partner study data under the hypothesis of constant infectivity can, therefore, be highly misleading at a public health level, particularly when extrapolated to multiple casual contacts.