Dramatic rises in youth violence in the United States beginning in the 1980s coupled with high visibility acts such as school shootings have resulted in a corresponding proliferation of programs designed to prevent aggression and violence in children and youth. Parallel with this increasing programmatic expansion, there have been repeated calls for rigorous evaluations of programs and identification of "best practices" that merit dissemination and implementation. Still, in spite of extensive efforts to document effective practices in youth violence prevention, there remains a somewhat confusing array of evidence for effectiveness and lack of effectiveness, as well as a broad range of interpretations of this evidence. From both a program and a policy perspective, it is at best difficult to sift through this seemingly contradictory assortment of evidence. Perhaps the key to unlocking this array of findings and interpretations is to rethink the very questions addressed. Asking "what works" in youth violence prevention suggests a simple, dichotomous answer (that is, works or does not work) about efforts to prevent or modify a problem behavior marked by its complexity and the multiplicity of associated risk and protective mechanisms. Even under the best circumstances it is unlikely that a violence prevention program "worked" for all: some participants may have improved, some may have stayed the same, and some may have increased their aggression, as has been noted in recent examinations of iatrogenic or negative effects interventions. In this chapter, the authors suggest that a more accurate understanding of effectiveness as well as failure can be found by reframing the questions and corresponding evaluation designs. Rather than asking, "What works in youth violence prevention?" the authors suggest that it is most important to ask: (a) what works and what does not work; (b) for whom; and (c) under what conditions? The authors discuss each component in more detail, providing illustrative examples from a large-scale prevention study, the Metropolitan Area Child Study (MACS). As they point out, this study provides an example of a multi-component, multi-context intervention that was more effective for some children and less effective for others.