The objectives of the study are to report the 20-week base rate of violent behaviour in the community among a general psychiatric patient population from Stockholm and to establish the validity of a violence risk assessment software program, Classification of Violence Risk (COVR), in a European setting. Three hundred and thirty one patients at two psychiatric hospitals in Stockholm were interviewed upon discharge. Telephone interviews with the patients and collaterals were conducted 10 and 20 weeks later. The violent behaviour was also measured through a national criminal register. The allocation of patients into different risk groups according to COVR software was compared with the occurrence of actual acts of violence during the follow-up. The base rate of violent behaviour was 5.7% and a ROC-analysis showed that the AUC for COVR was 0.77. Since there were few patients in the high risk groups, the 95% confidence interval for the proportion of violent patients was wide. The base rate of violent behaviour is relatively low in Sweden and prediction is therefore difficult. The predictive validity of COVR software is comparable to other risk assessment tools.