Results are reported of a quantitative synthesis of 70 published evaluations of educational programs for people with long-term health problems and regimens that include drugs. Study methods and interventions are rated according to methodological and educational principles, respectively. The differential effects achieved by seven types of educational techniques are compared. The largest effect-size values (in standard deviation units) for knowledge were in three empirically distinct and homogeneous groups of studies--those testing one-to-one counseling, group education, and one or both of these strategies in combination with audiovisual materials (range = 0.73-1.13). Homogeneous subgroups of behavioral studies (grouped according to type of intervention) were found within the studies measuring drug utilization errors, but the overall group was homogeneous. The absence of significant differences among the subgroups precluded subgroup comparison. The overall effect size was 0.37, indicating substantially decreased drug errors. Analyses of study and other intervention characteristics found the educational rating score to be the strongest predictor of effect size for both knowledge and drug errors. Patient package inserts were predictors of lower effect sizes.