OBJECTIVE: Evaluation of patients' potential for violence is an important component of care in psychiatric emergency and inpatient settings. Although situational variables are widely acknowledged to influence the risk of violence by psychiatric patients, most past research has been limited to patient attributes and has neglected the interpersonal context in which violence occurs. To the authors' knowledge, this is the first study to assess a new type of situational risk factor, the quality of the initial therapeutic alliance between the therapist and patient, as a predictor of the risk of violent behavior during short-term hospitalization. METHOD: The admitting physician's written initial evaluation for each of 328 patients hospitalized on a locked inpatient unit was reviewed by using a standardized alliance scale. The checklist measures the patient's level of active collaboration with the treatment process. Acute symptoms were rated at admission by physicians using the Brief Psychiatric Rating Scale. Nurses rated aggressive behavior in the hospital with the Overt Aggression Scale. RESULTS: Patients who had a poorer therapeutic alliance at the time of admission were significantly more likely to display violent behavior during hospitalization. Logistic regression analysis showed that the quality of the initial therapeutic alliance remained a strong predictor of violence even when other clinical and demographic correlates of violence were considered concurrently. CONCLUSIONS: The results suggest a new class of situational, interactional variable--reflected in the quality of the therapeutic relationship--that may be useful in evaluating patients' potential for violence. Implications for risk management are discussed.