Forensic psychologists and psychiatrists were shown case summaries of patients hospitalized with mental disorders and were asked to judge the likelihood of the patient harming someone within 6 months after discharge from the hospital. They also judged whether the patient posed a high risk, a medium risk, or a low risk of harming someone after discharge. Questionnaires were mailed to all 1,487 members of the American Academy of Psychiatry and Law. Completed questionnaires were received from 409 individuals, for a response rate of 28 percent. Providing clinicians with response scales allowing more discrimination among smaller probabilities led patients to be judged as posing lower probabilities of committing harmful acts. This format effect was not eliminated by having clinicians judge relative frequencies rather than probabilities or by providing them with instruction in how to make these types of judgments. In addition, frequency scales led to lower mean likelihood judgments than did probability scales. However, at any given level of likelihood, a patient was judged as posing a higher risk if that likelihood was derived from a frequency scale than if it was derived from a probability scale. Similarly, communicating a patient's dangerousness as a relative frequency led to much higher perceived risk than did communicating a comparable probability. Different reactions to probability and frequency formats appeared to be attributable to the more frightening images evoked by frequencies. Implications for risk assessment and risk communication are discussed. 50 references, 9 tables, and 5 figures