A RANDOMIZED FIELD EXPERIMENT IS DESCRIBED AS ONE IN WHICH HALF OF AN ELIGIBLE TARGET GROUP IS ASSIGNED TO THE NEW PROGRAM AND HALF IS ASSIGNED, AT LEAST TEMPORARILY, TO A CONTROL CONDITION WHERE NORMAL SERVICES ARE RECEIVED. THE PURPOSE OF THE RANDOMIZED ASSIGNMENT IS TO GUARANTEE THAT THE PROGRAM PARTICIPANTS ARE EQUIVALENT TO NONPARTICIPANTS IN THE LONG RUN, AND CONSEQUENTLY, THAT COMPARISON OF THE TWO GROUPS WILL BE AS FAIR AND UNEQUIVOCAL AS POSSIBLE. ALTHOUGH RANDOMIZED TESTS ARE CONSIDERED TO BE THE MOST ACCURATE IN ESTIMATING PROGRAM COSTS AND EFFECTS, IT IS ACKNOWLEDGED THAT THEY MAY NOT ALWAYS BE APPROPRIATE OR FEASIBLE IN PRACTICE. THREE CRITERIA ARE DISCUSSED FOR GAUGING THE FEASIBILITY OF RANDOMIZED FIELD TESTS: PREVIOUS USE ESTABLISHING A PRECEDENT; PILOT TESTS AND PRELIMINARY TESTS AS A DEMONSTRATION OF FEASIBILTY; AND THE PROPER CONDITIONS FOR EXECUTING THE THE TESTS. MORAL ARGUMENTS AGAINST RANDOMIZATION ON THE BASIS OF THE EXCLUSION OF THE CONTROL GROUP FROM POSSIBLE HELP ARE CONSIDERED. ALTERNATIVE TESTING METHODS ARE BRIEFLY DISCUSSED, AND THE SUFFICIENCY OF MANPOWER AS A DETERMINANT OF TESTING METHOD SELECTION IS CONSIDERED. (RCB)