Mathematical models suggest that multiple concurrent partnerships (MCP) may be the "key driver" of generalized HIV epidemics. Empirical studies, however, show that MCP are not more common in regions of Africa with high HIV prevalence than in regions with concentrated epidemics. This might be due to systematic under-reporting of MCP during population-based surveys of sexual behaviors (e.g., DHS). We use unique data from a sexual network study conducted on a small island of Lake Malawi to test this hypothesis. We elicited the sexual partners of 923 inhabitants of the island and traced these partners in existing rosters of the island's population. We compared self-reports of sexual relationships made by respondents to reports made by their partners. We then used this information to assess whether self-reported measures of MCP possibly underestimate the contribution of MCP to HIV spread in this context. We find that the prevalence of MCP at the time of the survey was significantly higher in this population according to estimates derived from network data. Using a simple model of HIV transmission in conjunction with network data, we show that the proportion of recent infections attributable to MCP is 50-60% higher than what self-reported data would imply. Our results suggest that new interventions addressing patterns of sexual networking and multiple partnerships at the population level may be beneficial.