One randomized experiment was conducted in Broward County, Fla. It involved the evaluation of a domestic violence counseling program that the Florida legislature mandated for persons convicted of domestic violence offenses. Although at first hesitant about implementing random assignment as the method for evaluating the counseling program's effectiveness, once the judges understood why other methods could not exclude competing explanations for outcomes, they made a commitment to the experiment. As word of the proposed research study spread, only the prosecutor's office opposed the study as being unethical, because it excluded some offenders from the program, thus putting potential victims at risk. The study continued to move forward despite the prosecutor's commitment to appealing the judges' decision. Researchers did not know from one day to the next, however, whether or not the experiment would be shut down. For 17 months, the researchers were faced with the fact that they had little control over external events that could greatly affect the status of their experiment. The goal of the second field experiment in Portland, Ore., was to determine whether arrest for misdemeanor domestic violence, followed by the prosecution of the offenders and the empowerment of the victim, was more effective in reducing violence between heterosexual married or unmarried intimates than arrest by itself. Problems encountered in both experiments were gaining and maintaining cooperation from agencies involved in the experiment, problems in sample size due to overestimates in the size of the eligible population, resistance and/or hostility to experiments by those within the affected agencies, hostility engendered outside the affected agencies, and high staff turnover caused largely by such hostility. The article concludes with a discussion of what researchers can do to forestall such problems. 35 references