Recent reviewers of well-controlled driver improvement program evaluations have suggested that some programs result in measurable reductions in violations but not crashes. A comprehensive review of 65 studies evaluating driver improvement activities was conducted to determine the generality of these findings and to explore possible explanations of the lack of correspondence between violation and crash effects. Nineteen studies evaluating 59 driver improvement activities were found to be methodologically adequate. The major findings of the review are: (1) that driver improvement activities generally do result in reductions in violations; (2) there is an unpredictable and sometimes undesirable effect on crashes even in the presence of desirable violation effects; and (3) the lack of correspondence between violation and crash effects is not explained by lack of statistical power or by the types of violations affected. Further examination of driver improvement interventions revealed no strong evidence for different effects related to characteristics such as direct vs. indirect participant contact and group vs. individual contact.