These problems often prevent the use of an ideal research design, such as random assignment to a control group that receives no treatment. Such a design seems impossible to justify ethically, because dangerous behavior is being treatment. A comparison group design, in which men are randomly assigned to two different forms of treatment, is more ethically acceptable. However, the small number of men who enroll in treatment groups makes it difficult to have enough men in each treatment condition. In addition, when clients are referred from the criminal justice system, the use of experimental designs may be restricted. The battered woman is actually the primary client and should give informed consent before her partner takes part in an experiment. Efforts must also be made to reduce the duress factor, which can make it difficult for some victims and abusers to make decisions about participation in research projects. Providing long-term followup is another problem, particularly in terms of maintaining confidentiality. The measurement of outcomes is a further issue, involving decisions about who should report, what behavior should be measured, and whether substitute measures can be used. Future research should use multiple designs and multiple measures to permit conclusions about the effects of various treatment formats, methods, settings, and client types. 45 references.