Misleading findings will result if a comparative area-based crime-reduction trial is viewed as an RCT. Evaluations that compare areas, one with an intervention and one without, are neither randomized nor controlled; and the crime events counted are correlated rather than statistically independent. Statistical biases will result because of the lack of equivalence between areas at the outset. Obtaining reliable evidence for the effectiveness of a crime-reduction measure requires the random assignment of a number of areas to receive the intervention or be kept unchanged. The data must then be analyzed with "area" as the unit of analysis. Further, trials should be designed to determine the number and size of participating areas, so as to obtain adequate information at minimum cost. This article compares valid evaluations of crime-reduction interventions with the method of health care research, which ensures the use of the correct unit of observation and analysis, incorporates randomization, and relies on statistical independence between observations. 3 tables, 2 figures, and 6 references