Psychosocial variables that predict dietary behavior become important targets for change in nutrition education programs. Psychosocial variables in models with higher predictability provide more effective levers to promote healthy dietary change. A review of the literature on models with psychosocial variables predicting dietary fat and fruit and vegetable consumption revealed generally low predictiveness, R2 < 0.3 (where R2 is the squared multiple correlation of the statistical model). No single theory provided models that regularly out-predicted others. When models predicted narrower categories of behavior (e.g. milk or salad consumption), predictiveness tended to be higher. Substantial problems were revealed in the psychometrics of both the independent and dependent variables. Little theory-based research has been conducted with adolescents, and the few studies done with children had low predictiveness. In order to increase the predictiveness of models, future research should combine variables from several theories, attend to the psychometrics of all variables, and incorporate variables that moderate the relationship of psychosocial to dietary behavior (e.g. genetics of taste, stage in the life course). Refinements on current research would include longitudinal designs and use of non-self-report methods of dietary behavior to supplement the self-report methods. Improved understanding of dietary behavior should lead to more effective dietary behavior change interventions.