Numerous studies have demonstrated the existence of regional variations in medical services for apparently similar populations. Often the intervention rates at vary by a factor of 3 or more, sometimes by as much as a factor of 10 or more (e.g., tonsillectomy, back surgery). This study seeks to understand why these variations exist and persist, i.e., the microfoundations"""""""" of variations. The study follows the model in Phelps and Parente (1988), where doctors hold different beliefs about the efficacy of care or other important parameters affecting the decision to treat. This study then determines the relative importance of various sources of information to doctors about such parameters, by first calculating individual doctors' propensity to use medical care, and then by correlating those propensities with observable features of the doctor. This study first estimates the propensities of individual physicians to use medical care along different dimensions (propensity to hospitalize, refer to specialists, length of stay upon hospitalization, cost per hospitalization, use of ancillary tests, etc.). It then regresses the propensities of each doctor on identifiable characteristics, such as type and extent of training, location of training, age, and duration in the community. The study determines the relative importance of various sources of information to individual physicians that they might use in recommending or making treatment choices for their patients. This study uses data from an IPA (210,00 members) wherein each patient has a primary care physician responsible for their care, and without whose direction no hospitalization or referral to specialty care may take place. This allows the calculation of rates of hospitalization, referral, etc. in ways previously available only using geographic regions to determine the relevant population. The study will also extend previous studies on aggregate variations analysis. These studies assist in setting priorities for medical practice and technology assessment, and deserve further refinement. The aggregate data also provide a test the information-cost model by assessing correlations of utilization across regions for procedures within medical specialty; small or null correlations are not consistent with standard market models or induced demand models.