Observational studies of demand for mental health services showed much greater use by those with more generous insurance, but this difference may have been due to adverse selection, rather than in response to price. This paper avoids the adverse selection problem by using data from a randomized trial, the RAND Health Insurance Experiment (HIE). Participating families were randomly assigned to insurance plans that either provided free care or were a mixture of first dollar coinsurance and free care after a cap on out-of-pocket spending was reached. We estimate that separate effects of coinsurance and the cap on the demand for episodes of outpatient mental health services. We find that outpatient mental health use is more responsive to price than is outpatient medical use, but not as responsive as most observational studies have indicated. Those with no insurance coverage would spend about one-quarter as much on mental health care as they would with free care. Coinsurance reduces the number of episodes of treatment, but has only a small effect on the duration and intensity of use within episodes. Users appear to anticipate exceeding the cap, and spend at more than the free rate after they do so.