This paper reviews experimental methods for the study of the responses of people to violence in digital media, and in particular considers the issues of internal validity and ecological validity or generalisability of results to events in the real world. Experimental methods typically involve a significant level of abstraction from reality, with participants required to carry out tasks that are far removed from violence in real life, and hence their ecological validity is questionable. On the other hand studies based on fi eld data, while having ecological validity, cannot control multiple confounding variables that may have an impact on observed results, so that their internal validity is questionable. It is argued that immersive virtual reality may provide a unifi cation of these two approaches. Since people tend to respond realistically to situations and events that occur in virtual reality, and since virtual reality simulations can be completely controlled for experimental purposes, studies of responses to violence within virtual reality are likely to have both ecological and internal validity. This depends on a property that we call"plausibility"- including the fi delity of the depicted situation with prior knowledge and expectations. We illustrate this with data from a previously published experiment, a virtual reprise of Stanley Milgram"s 1960s obedience experiment, and also with pilot data from a new study being developed that looks at bystander responses to violent incidents.