Individual differences in physicians' laboratory use and referral rates are important aspects of practice variation that have real financial and health consequences. A way to explain these differences is needed. In this empirical study, physicians' risk attitudes (measured on a multidimensional scale) are shown to be good predictors of use rates for certain specific laboratory procedures, but not good predictors of physicians' referral rates. A 15-item survey form that measured risk-taking attitudes in the financial, health, social, and ethical domains was administered to all clinical faculty at an academic family practice center (n = 14). Each physician's utilization rates for the 17 most frequently ordered laboratory procedures were calculated for all patient visits for one calendar year. Overall referral rates were calculated for the same period. Physicians' risk attitudes (12 completed the survey) accounted for over 50% of the variance for several of the laboratory procedures. For example, the rank-order correlation between the complete blood count utilization rate and a Likert-scale item measuring physicians' propensity to take physical risks was 0.91 (p less than 0.001). The details of these findings help to explain an important component of practice variation.