In the context of a food assistance program in rural Haiti, we developed measures of the effectiveness of community health worker (CHW)-delivered behavior change communication (BCC). We administered knowledge tests to 954 mothers and 38 CHWs to define 4 measures: CHW knowledge, maternal knowledge, knowledge-sharing efficacy (proportion of CHW knowledge shared), and shared correct knowledge between the CHWs and the mothers with whom they interacted. On the basis of the tests, CHWs had high knowledge (93% correct), mothers scored 72% on maternal knowledge, the proportion of CHW knowledge shared was 75%, and shared correct knowledge between CHWs and mothers was 70%. Factors affecting maternal knowledge included CHW characteristics (unmarried: β = -0.070, P < 0.05), long duration (19-45 mo) of program participation (β = 0.034, P < 0.05), and having multiple sources of health and nutrition information (β = 0.072, P < 0.01). Shared correct knowledge and CHW knowledge-sharing efficacy were positively associated with CHW (age, education) and program participation characteristics. We parsed the relative contributions of CHW characteristics to total and proportion of shared CHW knowledge. We observed a positive association between CHW education and shared correct knowledge between the CHWs and mothers (β = 0.328, P < 0.01), not because of more knowledge (β = -0.012, P > 0.05) but rather because of greater knowledge-sharing efficacy (β = 0.340, P < 0.01). These measures provide a means of examining features of program delivery and interpreting impact evaluation results. They show that BCC outcomes are conditional on both frontline workers' knowledge and knowledge-sharing efficacy. Whereas most programs focus on content training to improve CHWs' knowledge, it is also important to strengthen process training and support to foster knowledge-sharing efficacy.