The analysis focused on the deterrent effect of arrest in cases of domestic violence in terms of preventing further attacks. The findings indicated that arrest works better for some suspects than for others, but did not support the notion that a particular arrest will necessarily lead to a beneficial outcome. The usefulness of the interaction effects to the criminal justice apparatus is unclear unless police officers can easily and accurately determine the good risks at the scene and the legal system allows sanctions to vary by the degree of risk posed by a suspect. A better way to protect the victims of domestic assault would be to focus on the policies applied after taking a suspect into custody, i.e., setting very high bail for bad-risk suspects and helping the victim obtain shelter or a court protection order. Finally, the existence of interaction effects implies that the deterrence value of arrest averaged over all suspects in a given area varies depending on the mix of suspects and local police practices. 4 tables, 14 notes, and 26 references