Motivated by the Green and Porter (1984) and Rotemberg and Saloner (1986) models, we construct lab experiments to test the effects of two factors on collusion: information (regarding next period's demand state) and monitoring (of a rival's past action). Results indicate that information may facilitate collusion more than monitoring, especially as subjects gain experience. A robust finding is that subjects in the Rotemberg and Saloner treatment cooperate as predicted by this theory: collusion falls dramatically in anticipation of unusually large demand and returns to high levels otherwise. These results suggest that tacit and fairly elaborate collusion could arise in stochastic environments.