An experiment is reported for payoff-equivalent public good and common pool games with high caste and low caste West Bengali villagers. Tests are reported for models of unconditional social preferences, models of reciprocity, and cultural identity. Results from the artefactual field experiment indicate that when information about caste is withheld no significant difference is observed in the efficiency of play between the villagers and student subjects at American universities in games with positive and negative externalities. In contrast, making the hereditary class structure salient induces different behavior among villagers. Providing caste information leads to: (i) the lowest level of efficiency when low caste first movers interact with a low caste second mover, and (ii) the highest level of efficiency when high caste first movers interact with a high caste second mover. Cross-caste play generates intermediate levels of efficiency.