The study retrospectively collected data that documented the types of resources available to victims of domestic violence since 1976 and examined their relationship to the changing patterns of partner homicide. The analysis was based on a panel data set of 48 of the 50 largest U.S. cities for the years 1976 to 1996. The researchers estimated separate panel models for eight possible combinations of victim sex, race, and marital relationship. The analysis incorporated 11 indicators of the State and local domestic violence resources. Four were measures of State statutes; five measured components of local police and prosecution policy; and two measured the strength of legal advocacy programs and the prevalence of hotlines in the city. The analysis controlled for non-intimate adult homicide rates, marriage and divorce rates, women's relative educational attainment, and welfare benefit levels in each of the cities. Poisson regression models were used to estimate the effects of these variables on homicide. Additional methodology was designed to reduce the chances that the estimates were statistical artifacts due to unusually influential cities or years. The study provided mixed support for the general exposure reduction hypotheses. A little more than half of the findings support the predictions of exposure reduction, and the others show that domestic violence resources are associated with more killings for some victim types. This backlash effect was especially pronounced for unmarried partners. The adoption of a warrantless arrest law was associated with fewer killings of white women and black unmarried men. Increases in the willingness of prosecutors' offices to take cases of protection order violation were associated with increases in the homicide of white married intimates, black unmarried intimates, and white unmarried females. Also, an untoward consequence of cutting AFDC (Aid for Families With Dependent Children) payment levels apparently has been increased homicide victimization of black married men, black unmarried partners, and white unmarried females. Other resources and conflicting results depended on victim category. Evidence of increased lethality, and even the null findings, could reflect failures of the criminal justice and social service systems to protect victims adequately once they accessed their services. More research is needed to better understand the dynamics of successful exposure reduction compared to unsuccessful cases, so that policymakers and practitioners can tailor resources to the diverse needs of a heterogeneous population of women and thus reduce prevention failures. 4 figures, 2 tables, and 57 references