Every year thousands of firms are engaged in research joint ventures (RJV), where all knowledge gained through R&D is shared among members. Most of the empirical literature assumes members are non-cooperative in the product market. But many RJV members are rivals leaving open the possibility that firms may form RJVs to facilitate collusion. We examine this by exploiting variation in RJV formation generated by a policy change that affects the collusive benefits but not the research synergies associated with a RJV. We use data on RJVs formed between 1986 and 2001 together with firm-level information from Compustat to estimate a RJV participation equation. After correcting for the endogeneity of R&D and controlling for RJV characteristics and firm attributes, we find the decision to join is impacted by the policy change. We also find the magnitude is significant: the policy change resulted in an average drop in the probability of joining a RJV of 34% among telecommunications firms, 33% among computer and semiconductor manufacturers, and 27% among petroleum refining firms. Our results are consistent with research joint ventures serving a collusive function.