We show how experimental results can be generalized across diverse populations by leveraging knowledge of mechanisms that produce the outcome of interest. We use Structural Causal Models (SCM) and a refined version of selection diagrams to represent such knowledge , and to decide whether it entails conditions that enable generalizations. We further provide bounds for the target effect when some of these conditions are violated. We conclude by demonstrating that the structural account offers a more reliable way of analyzing generalization than positing counterfactual consequences of the actual mechanisms.