Starting from an agent-based interpretation of the well-known Bass innovation diffusion model, we perform a Montecarlo analysis of the performance of a method of simulated_x000d_ moment estimator. We show that nonlinearities of the moments lead to a small bias in the estimates in small populations, and prove that our estimates are consistent and converge to the true values as population size increases. Our approach can be generalized to the estimation of more complex agent-based models. However, a trade-off emerges between model inadequacy and data inadequacy. This is particularly severe when only aggregate information is available, as common with diffusion data.