THE PROCESS OF RANDOMIZATION IS ESSENTIAL TO ANY EVALUATION APPROACH WHICH SEEKS TO ASSESS POLICIES OR PROGRAMS THROUGH AN EXPERIMENTAL DESIGN. ALTHOUGH THIS PROCESS IS HARD TO USE OUTSIDE CONTROLLED LABORATORY SETTINGS, A HIGHLY SUCCESSFUL RANDOMIZATION PROCEDURE WAS DEVELOPED FOR EVALUATION OF 3 COMMUNITY-BASED PROGRAMS DESIGNED TO PROVIDE SERVICES TO EX-OFFENDERS IN THE CHICAGO, ILL., METROPOLITAN AREA. THE FACT THAT CLIENTS WERE BEING TURNED AWAY FROM SERVICES BECAUSE OF LACK OF MANPOWER SUPPORTED THE ARGUMENT THAT A RANDOMIZATION PROCEDURE WAS THE MOST EFFICIENT WAY TO ASSIGN CLIENTS. IN ORDER TO REMAIN AS CLOSE AS POSSIBLE TO NORMAL OPERATING PROCEDURES, CLIENTS HAD TO BE ASSIGNED TO ONE OF SEVERAL TREATMENT ALTERNATIVES ON EITHER A DAILY OR WEEKLY BASIS. A UNIQUE RANDOM ASSIGNMENT PLAN WAS USED EACH WEEK, AS THERE WERE FACTORS HINDERING THE RANDOM ASSIGNMENT OF CLIENTS, SUCH AS THE FACT THAT THE NUMBER OF CLIENTS ENTERING THE PROGRAMS WAS NOT KNOWN IN ADVANCE AND VARIED FROM WEEK TO WEEK, EACH PROGRAM COULD ONLY HANDLE A CERTAIN NUMBER OF CLIENTS, AND THE ASSIGNMENT OF CLIENTS WAS NOT MADE BY THE PERSONS HANDLING THE EVALUATION. THE PROCEDURE ADEQUATELY HANDLED AN ASSIGNMENT PROCESS FOR CLIENTS WHO TRICKLED IN AND WERE PERMITTED TO QUEUE UP FOR SMALL-BATCH ASSIGNMENTS. THE EXTENSION OF THESE PROCEDURES TO DAILY, OR EVEN HOURLY, TRICKLE ASSIGNMENTS IS STRAIGHTFORWARD. IN INSTANCES WHERE REAL-TIME TELEPROCESSING IS AVAILABLE, SLIGHT MODIFICATIONS IN SOFTWARE COULD BE PERFORMED IN ORDER TO PRODUCE A UNIQUE ASSIGNMENT PLAN FOR ANY NUMBER OF INDIVIDUALS AND ANY DESIRED TIME INTERVAL. ILLUSTRATIONS OF THE EXPERIMENTAL DESIGN, TABLES SHOWING THE ALGORITHMIC FORMULA AND THE ASSIGNMENT PROCEDURES, AND REFERENCES ARE PROVIDED. (KCP)