This Article analyzes data from the Colorado Springs Spouse Abuse Experiment. In that experiment, suspects apprehended for misdemeanor spouse abuse were assigned at random to one of four treatments: (1) an emergency order of protection for the victim coupled with arrest of the suspect; (2) an emergency order of protection for the victim coupled with immediate crisis counseling for the suspect; (3) an emergency order of protection only; or (4) restoring order at the scene with no emergency order of protection. Outcome measures are taken from official police data and from follow-up interviews with victims. Using Bayesian procedures to take previous experiments into account, the balance of evidence supports a deterrent effect for arrest among 'good risk' offenders, who presumably have a lot to lose by being arrested. The balance of evidence is far more equivocal for a 'labeling effect' in which an arrest increases the likelihood of new violence.