Trust is thought to be an important driver of economic growth and other economic outcomes. Previous studies suggest that the decision to trust is driven by a combination of risk attitudes, distributional preferences, betrayal aversion, and beliefs about the probability of being reciprocated. We compare the results of a binary trust game to the results of a series of control treatments that remove the effect of one or more of these components of trust by design. This allows us to decompose variation in trust behavior into its underlying factors. Our results imply that beliefs are the main driver of trust, and that the additional components only play a role when beliefs about reciprocity are sufficiently optimistic. Our decomposition approach can be applied to other settings where multiple factors that are not independent amongst each other affect behavior. We analyze its advantages over the more traditional approach of including measures of relevant factors derived from separate tasks, in particular with respect to measurement error and omitted-variable bias.