The reliability of clinical predictions of dangerous behavior in mentally ill offenders is questioned because most of the data on the accuracy of such predictions concerned mental patients released -- contrary to psychiatric advice -- from maximum security facilities by court decisions. A study conducted at the Patuxent Mental Institute for Defective Deliquents in Maryland failed to reveal any ability on the part of the staff to accurately identify potentially dangerous inmates. Tabular data are included concerning clinical and statistical dangerousness predictions by various researchers based on false-positive rates. Accuracy of clinical predictions of dangerous behavior rarely exceeded in predictions obtainable simply by chance, even among groups generally considered high risk. Tabular data are also provided on the accuracy of clinical and statistical predictions of patients, assaultiveness while hospitalized; on the accuracy of clinical and statistical analysis in predicting assaultiveness in the community, and subsequent arrest for murder, manslaughter, or assault. The apparent psychiatric inabilities in predicting dangerous behavior have important implications for due process protection in both the mental health and the criminal justice system. Forty references are appended.