OBJECTIVE. We compared the validity of hospital admission and length of stay reports from patients, outpatient providers, and hospitals, and we examined possible sources of error. DATA SOURCES. Data were collected from people enrolled in a randomized trial of treatment for severe mental illness and substance use disorders, from community mental health centers (CMHCs), and from hospitals. Reports for each of the 74 study participants covered two-year time periods beginning and ending at various times between 1989 and 1993. STUDY DESIGN. We compared reports from the various sources and constructed a hybrid with data from all three sources. Using parametric and non-parametric statistics, we compared patient, CMHC, and hospital reports with each other and with the hybrid source. In subsequent regression analyses we explored correlates of reporting accuracy. PRINCIPAL FINDINGS. Single-source reports underestimated hospital use, but when patient and CMHC reports were combined, results were very similar to those obtained by the more laborious hybrid method. Patient reports became less accurate as the time between discharge and reporting increased; people with bipolar disorders reported admissions with greater accuracy than did people with schizophrenia. CMHC reporting accuracy decreased as the distance to the admitting hospital increased and were less accurate for people with more severe psychiatric symptoms. CONCLUSIONS. Reports from single sources are likely to underestimate hospital use for different reasons. Combining carefully collected data from patients and outpatient providers produces estimates of hospital use that are substantially the same as those developed through methods that are more laborious and costly.