Spurred by its successful delivery of credit to poor borrowers in diverse areas of the developing world, joint liability lending has caught the imagination of development theorists and practitioners. Various theories have arisen to explain why joint liability group-based lending can be an improvement over traditional individual-based lending. We test across four representative and oft-cited models from three strands of these theories, focusing primarily on their implications for group repayment rates given that the group contract is being used. The basis for comparison is two-fold. We evaluate how well the focal variables of each model explain the group's probability of repayment. We also develop the implications of these models over a common set of variables, extending them wherever reasonable, and use the divergent implications to distinguish between the models. These extensions yield some theoretical predictions that appear interesting in their own right. We find that correlation of borrower outputs can have a positive impact on repayment rates in a range of models. Cooperation in the group can have a negative effect on repayment rates by reducing the use of cheap punishments. We test across the models using survey data from 262 Thai joint liability groups of the Bank for Agriculture and Agricultural Cooperatives (BAAC) in conjunction with extensive survey data from 2880 villagers of the same villages. The Besley and Coate model of strategic default is strongly supported. The Banerjee, Besley, and Guinnane model of moral hazard and monitoring is supported in its predictions regarding monitoring, but not repayment. Moderate support is found for the Stiglitz model of moral hazard and cooperation and the Ghatak model of adverse selection. There is also a significant regional flavor to the results, as the Besley and Coate model seems especially accurate in the poorer northeast region farther from Bangkok, and the Ghatak and Stiglitz models perform relatively well in the central region. We also review and compare our results to the existing empirical literature, finding common ground, but also new results in several areas. We seem to be among the first to include the interest rate and degree of joint liability in regressions, to measure covariance more or less directly and find it a positive factor, and to find significant effects of social sanctions on repayment.