I study the effects of integration among generalist and specialist physicians on referral and treatment choices, inpatient admissions, and the cost of medical care. I first construct a novel data set that links the structure of physician organizations to physician behavior for two groups of individuals deemed most likely to benefit from generalist-specialist integra- tion: individuals diagnosed with hypertension and diabetes mellitus. Due to differential selection across integrated and non-integrated practices, ordinary least squares regression is unlikely to identify the effects of interest. To surmount the identification problem, I simulate random assignment of beneficiaries to physician organizations in an instrumental variables framework. My estimates confirm that integration among generalist and special- ists significantly affects referral decisions and treatment patterns. I fail to detect evidence that this kind of integration improves medical outcomes or reduces medical care spend- ing. These findings strongly suggest that policies increasing generalist-specialist integration in piece-rate or fee-for-service environments are unlikely to achieve their goals. Finally, I consider the theoretical literature that might rationalize generalist-specialist integration patterns. I show that most potential explanations are inconsistent with my evidence, save for David and Neuman (2011), suggesting that task adhesion should be a focal point for future research.