While much research has been written about physician-preferred learning methods and self-directed learning, no published study describes the physician's self-selected learning methods specific to problems arising in his practice. Study subjects were 366 primary care physicians asked to describe a recent clinical problem for which they needed more knowledge or skill to solve. For the specific problem of his/her own patient, each physician was asked how he/she gained the specific ability to solve the problem. Fifty-five learning methods were described by the study population and were categorized as three outcome variables: reading and related activity, 47.8% of study subject responses, formal and informal consultation and/or referral (33.3%), and formal continuing medical education (CME) (18.9%). Analysis for risk similarities led to progressive recombination of 64 problem categories to two predictive variables: clinical problems requiring more knowledge and clinical problems requiring more technical or communication skill to solve. Multiple logistic regression was carried out. The final predictive model revealed that of the factors studied, problem type was the major determinant of self-selected learning method in the practice setting. While reading is the most frequently selected method overall, physicians are more likely to select formal CME when the problem requires more technical expertise or communication ability to solve (odds ratio = 2.31, confidence interval = 1.25–4.25). Region and year of graduation added only slightly to the model. Gender, certification, and size of town did not add to the predictive model. CME providers must be aware of physicians' practice-based choices of learning method and incorporate this into their CME activities.