We provide a direct test of the role of social preferences in voluntary cooperation. We elicit individuals' cooperation preference in one experiment and make a point prediction about the contribution to a repeated public good. This allows for a novel test as to whether there are "types" of players who behave consistently with their elicited preferences. We find clear-cut evidence for the existence of "types". People who express free rider preferences show the most systematic deviation from the predicted contributions, because they contribute in the first half of the experiment. We also show that the interaction of heterogeneous types explains a large part of the dynamics of free riding.