Accurate evaluations of the dangers posed by psychiatric inpatients are necessary, although a number of studies have questioned the accuracy of violence prediction. In this prospective study, we evaluated several variables in the prediction of violence in 63 inpatients with a DSM-IV diagnosis of schizophrenia or schizoaf-fective disorder. Nurses rated violent incidents with the Overt Aggression Scale. During hospitalization, sociodemographic variables, clinical history, neurological soft signs, community alcohol or drug abuse, and electroencephalographic abnormalities did not differ between violent and nonviolent groups. Violent patients had significantly more positive symptoms as measured by the Positive and Negative Syndrome Scale (PANSS), higher scores on the PANSS general psychopathology scale, and less insight in the different constructs assessed. A logistic regression was performed to discriminate between violent and nonviolent patients. Three variables entered the model: insight into symptoms, PANSS general psychopathology score, and violence in the previous week. The actuarial model correctly classified 84.13 percent of the sample; this result is significantly better than chance for the base rate of violence in this study. At hospital admission, clinical rather than sociodemographic variables were more predictive of violence. This finding has practical importance because clinical symptoms are amenable to therapeutic approaches. This study is the first to demonstrate that insight into psychotic symptoms is a predictor of violence in inpatients with schizophrenia.