This dissertation was motivated by the limited research on macro-level predictors of patient assessments of care. A theoretical model for studying multilevel predictors of patient assessments of care was developed, based on Donabedian’s structure-process-outcomes model and resource dependency theory, to guide this study. Overall, the sample included 24,887 medical/surgical patients from 173 hospitals, nested in 46 counties in CA; the data were derived from the 2002 NRC Picker Patients’ Evaluation of Performance in California (PEP-C) survey, the American Hospital Association (AHA) Annual Survey, and the Area Resource File (ARF). The study employed three-level hierarchical linear models, where patient, organizational (hospital), and market predictors were introduced in a sequential model-building approach to explain variations across data levels in process quality and overall satisfaction with care. Six of the ten hypotheses were either completely or partially supported by the results, providing support for the theoretical model. This study found that variations in all domains of patient assessments of care abound. Strikingly, however, most of the overall variations (95% - 99%) were within- hospitals, rather than between-hospitals or between-markets. Patient characteristics accounted for up to 13% of variations at the within-hospital level. Interestingly, despite the relatively small between-hospital and between-market variations, net of patient characteristics, organizational- and market-level characteristics predicted a sizable amount of the true variations in process quality and overall satisfaction with care domains. The findings of this study suggest that most of the variations in patient assessments exist within the patient-health-care-provider relationship. Overall, this study supports the robustness of patient assessments of care in elucidating and explaining sources of variations across data levels. Other implications are discussed.