In 1994, New York State enacted a law that mandates an arrest when an offense involving family members rises to the level of a felony or an order of protection is violated. Data from site 1 and site 2 were based on all qualifying domestic-violence incidents reported between March 1 and September 30, 1996; and data collection in site 3 (the smallest city) extended through August 31, 1997. An incident qualified if it involved sexual intimates and physical aggression and originated as a call for service. The incident was coded as an arrest if it resulted in a warrantless arrest or the issuance of either a warrant or criminal summons. Seven rational-system (legalistic) variables and 13 social structural variables were used in the multivariate analysis. Approximately 85 percent of the incidents in all three sites involved a male suspect and female victim, and minorities (Non-Whites and Hispanics) were overrepresented in all three sites. Logistic regression was used to identify predictors of arrest decisions. The findings indicate that case characteristics related to offense seriousness were predictive of arrest decisions when the suspect was present at the scene; however, the absence of the suspect at the scene significantly reduced the probability of an arrest and attenuated the relationship between traditional measures of offense seriousness and the arrest decision. This suggests that departments must develop procedures to pursue domestic-violence suspects without reliance on the initiative of victims. 5 tables, 8 notes, and 37 references