The emergent realist perspective on evaluation is instructive in the quest to use theory-informed evaluations to reduce health inequities. This perspective suggests that in addition to knowing whether a program works, it is imperative to know 'what works for whom in what circumstances and in what respects, and how?' (Pawson & Tilley, 1997). This addresses the important issue of heterogeneity of effect, in other words, that programs have different effects for different people, potentially even exacerbating inequities and worsening the situation of marginalized groups. But in addition, the realist perspective implies that a program may not only have a greater or lesser effect, but even for the same effect, it may work by way of a different mechanism, about which we must theorize, for different groups. For this reason, theory, and theory-based evaluations are critical to health equity. We present here three examples of evaluations with a focus on program theories and their links to inequalities. All three examples illustrate the importance of theory-based evaluations in reducing health inequities. We offer these examples from a wide variety of settings to illustrate that the problem of which we write is not an exception to usual practice. The 'Housing First' model of supportive housing for people with severe mental illness is based on a theory of the role of housing in living with mental illness that has a number of elements that directly contradict the theory underlying the dominant model. Multisectoral action theories form the basis for the second example on Venezuela's revolutionary national Barrio Adentro health improvement program. Finally, decriminalization of prostitution and related health and safety policies in New Zealand illustrate how evaluations can play an important role in both refining the theory and contributing to improved policy interventions to address inequalities. The theoretically driven and transformative nature of these interventions create special demands for the use of theory in evaluations (Authors).