We report an experiment on payoff-equivalent provision and appropriation games with high-caste and low-caste Indian villagers. A central question is whether caste identities affect resolution of social dilemmas. Making caste salient in the experiment elicits striking changes in behavior compared to the baseline treatment with no information about others’ castes. Homogenous groups with high caste subjects are more successful in resolving social dilemmas than homogenous groups with low caste subjects. The success of mixed groups in resolving such dilemmas is somewhere between, which is inconsistent with in-group vs. out-group identity models. Absent salient information on caste, behavior is inconsistent with unconditional social preferences but as predicted by reciprocity