OBJECTIVES: To improve quality and persistency of anti-depressive therapy in primary care by a correspondence course for GPs focusing on interaction between patient and physician. METHODS: One hundred sixty-nine depressive patients in two groups were observed in a cohort study. GPs in the education group (84 patients) received training material on therapy of depression. A parallel control group of 85 depressive patients recruited by physicians not obtaining training material was observed to compare effectiveness. In both groups patients were on Citalopram treatment. Measures: demographics, prescriptions, Clinical Global Impression Scale (CGI) and Hamilton Depression Scale (HAM-D). Additionally, patients were asked to answer questions concerning their treatment satisfaction and quality of life (SF-36). They were observed over a 6-month period. For external comparison of therapy duration to non-study conditions, data from the prescription database MediPlus (IMS HEALTH) were analysed. RESULTS: Comparing the two groups there were no differences in clinical data (CGI, HAM-D) and quality of life (SF-36). However, the following differences were found: In comparison to control group, patients in education group had more contacts to their GP. Especially in the beginning of therapy, there were shorter intervals between visits. Patients in education group received anti-depressive medication for a longer period. In education group the rate of discontinuing the pharmaceutical therapy within the recommended 6-month period was only 36% in comparison to 44% in the control group (reduction of 18%). When related to MediPlus data, the discontinuation rate was lowered by 46%. CONCLUSIONS: The IMPROVE-D training material for GPs improved therapy quality and persistency. The training resulted in more frequent physician contacts and careful drug prescription. Patients felt more bound to the physician's instructions and showed a better compliance. This resulted in a longer therapy duration which experts link to a reduced risk for relapsing.