A meta-analysis of the literature of controlled studies of educational and psychosocial interventions in the treatment of diabetes mellitus yielded 93 studies of 7451 patients testing the effects of eight intervention types: (1) didactic education, (2) enhanced education, (3) diet instruction, (4) exercise instruction, (5) self-monitoring instruction, (6) social learning/behavior modification, (7) counseling, and (8) relaxation training. An overall mean effect size (ES) of +0.51 +/- 0.11 was found moderate but significant (P less than 0.05) improvements for all intervention subjects. Physical outcome and knowledge gain were most affected, followed by psychological status and compliance. Diet instruction and social learning interventions showed the strongest (ES = +0.68 +/- 0.58 and ES = +0.57 +/- 0.42, respectively) and relaxation training the weakest (ES = +0.30 +/- 0.74) effects. Associations between study and sample characteristics and mean ES values were explored with type of setting and methodological weaknesses such as single group design and non-random assignment achieving statistical significance. Neither intervention type, number of visits, sex, age, nor type of diabetes were significantly correlated with mean ES values. Implications of these findings for clinical treatment and future research are discussed.