The study evaluated policies designed to protect battered women from suspects brought to prosecution following either a police officer's warrantless arrest or a victim's complaint directly to the prosecutor. Suspects included all men formally charged with misdemeanor assault against a female conjugal partner in Indianapolis between June 1986 and August 1987. The on-scene warrantless arrest (OSA) experiment involved 198 subjects arrested at the scene of a violent domestic disturbance. The victim-initiated complaint (VC) experiment involved 480 suspects identified by victim affidavits filed at the Prosecutor's Office. Based on randomized policy recommendations, prosecutors tracked cases toward one of three outcomes: pretrial diversion to rehabilitative counseling; adjudicated guilt with counseling as a probation condition; or other sentencing such as a fine, probation, or jail time. Victims were not permitted to drop charges under any of these policy tracks. Each VC defendant was also processed according to a randomized entry condition (summons or warrant) and a fourth randomized prosecution policy allowing the victim to drop charges. Following settlement in court, each case was monitored to determine if alternative criminal justice policies reduced repeat violence against the same victim. Data were obtained from victim and offender interviews and from official records. Outcome measures included prevalence, severity, and frequency of violence in each relationship and length of time until a new episode of violence following court settlement. Analysis of OSA and VC cases showed that policy alternatives to traditional sentencing were no more effective in protecting victims 6 months following case settlement. However, when VC defendants were arrested under a warrant and their victims were permitted to drop charges, women were significantly more likely to be safe from continuing violence. 54 references, 28 tables, and 7 figures