Since 1988 five evaluations of Federal demonstration programs in education have been implemented that had random-assignment designs to measure program impacts. The implementation of two of these evaluations, the evaluation of the School Dropout Demonstration Assistance Program and that of the Alternative Schools Random Assignment Program, are explored to lend support for several conclusions regarding random assignment designs. The first is that random assignment can be implemented in a variety of settings and at a scale that is suitable for measuring impacts precisely, but that it is generally poorly understood by educators. Consequently, it is difficult to implement without a great deal of discussion and negotiation. A second conclusion is that program staff fear that they will lose control of who is admitted to the program. The mechanics of random assignment need to be tailored to this concern. A final conclusion is that random assignment is most likely to fail when the pool of applicants is inadequate to support creation of a control group. When programs experience shortages of applicants, random assignment is not desirable. (SLD)