A computer was used to prospectively detect and suggest responses to simple, medication-related events as reflected in a computerized record in a prospective, randomized study of a diabetes clinic with primary-care responsibility. There were two categories of event/suggestions: when the last observation of a test required for medication control was too old, the computer suggested a repeat; and when an abnormality with therapeutic implications was detected, the computer suggested a specific change in therapeutics. Clinicians responded to 36% of events in the first category with computer reminders and 11% without (P less than 0.0001); they responded to 28% of events in the second category with computer assistance and 13% without (P less than 0.026). For the most clinically significant of these second category events, the difference was even greater: 47% with and 4% without computer assistance (P less than 0.0004). I believe that computer detection and response (in the form of reminders) to simple clinical events will change clinician behavior.