We examine whether the pattern of transfers among villagers can be explained either by a model of pure altruism or by a model of dynamic incentives, by comparing the predictions of these models to actual transfer patterns in Gambian villages. We find that the topology of transfer networks violate some of the predictions of both models. We then generalize the model of pure altruism to include caps on transfers that represent transaction costs, liquidity constraints, and social norms. Even with a parsimonious parametrization, the generalized model predicts equilibrium networks with topological features that match closely those of intra-village transfer networks in rural Gambia. Using a simulation-based indirect inference approach, we estimate the structural parameters of this model, relying only on income data and the topology of the transfer network for identification. We test for the existence of binding caps on transfers and find that network topologies are consistent with strong caps on transfers. Our structural estimates confirm that kinship is a strong predictor of altruistic preferences. We corroborate our structural estimates using proxies for transaction costs from surveys.