OBJECTIVE: To select and evaluate characteristics of primary care practice groups relevant for stratification prior to randomization. METHOD: Structured telephone interviews and pre- and post-intervention prescription data. SETTING: Additional study in an RCT to rationalize prescribing in primary care, addressing groups of pharmacists and doctors. Representatives of 61 primary care practice groups in the Netherlands. MAIN OUTCOME MEASURES: Identification and evaluation of primary care practice group characteristics related to changes in prescribing. These characteristics were evaluated by estimation of incidence rate ratios (Poisson regression). RESULTS: Of practice groups 40 representatives (66%) participated in our study. Three characteristics were found to be most relevant for stratification: the purpose of the practice groups (whether they participated in information exchange (n = 14) or binding consensus on pharmacotherapy (n = 26)), the use of a formulary (22 practice groups did and 18 groups did not) and the use of feedback data (22 groups did and 18 groups did not). These characteristics strongly modified the effect of the program on prescribing behaviour. While the overall effect of the program was to significantly reduce the prescribing of the targeted drugs, this change was not consistent in all strata. CONCLUSION: Assessment of the characteristics of practice groups made it possible to define factors to be used for a stratified randomization of practice groups which in retrospect indeed modified the effect of an educational intervention directed to change prescribing. Such pre-randomization assessments can be a useful tool in interventions designed to improve practice patterns in groups of physicians.