OBJECTIVE An overview of studies on predictors and on the accuracy of prediction of inpatient violence should be given. METHOD To date, the published data do not allow a systematic meta-analysis due to different sample characteristics, different measures and definitions of violence, and different time frames of observation. Published studies were reviewed regarding significant predictors of inpatient violence. RESULTS Predictors of violence in institutional settings are different from predictors of violence in the community: variables such as sex, age, diagnosis and alcohol abuse play a minor role, while clinical and psychopathological variables are prominent. Only history of violence is a robust static predictor. The total level of positive and general psychotic symptoms seems to enhance the violence risk of inpatients, whereas results concerning specific features like delusions or command hallucinations are contradictory due to inevitable problems of sample selection. The accuracy of clinical predictions is better than chance but limited by the effects of therapeutic interventions and research artefacts. CONCLUSION The author argues that more precise determinations of the violence risk in institutions will fail and that the role of environmental factors has often been underestimated.