CONCEPT FORMATION IN 50 KINDERGARTENERS WAS STUDIED BY REQUIRING THE CHILDREN TO SORT CARDS ACCORDING TO ONE OF FOUR ATTRIBUTES OF THREE DIFFERENT DIMENSIONS. THE OBJECTIVE WAS TO EXPLORE THE VALIDITY AND LIMITATIONS OF AN ALL-OR-NONE LEARNING MODEL FOR COMPLEX CLASSIFYING RESPONSES. INFORMATION WAS PRESENTED TO THE SUBJECT BY TWO POSITIVE EXAMPLES IN ONE PROBLEM SET AND BY A POSITIVE AND A NEGATIVE EXAMPLE IN THE OTHER SET. POSITIVE PROGRAM SETS WERE GIVEN FIRST. HALF OF THE CHILDREN BEGAN WITH GEOMETRICAL PROBLEM CARDS AND, AFTER REACHING CRITERION ON THESE, TRANSFERRED TO MORE DIFFICULT "PEOPLE" CARDS. THE OTHER HALF RECEIVED PROBLEMS IN REVERSE ORDER. WHEN POSITIVE EXAMPLES WERE GIVEN, POSITIVE TRANSFER EFFECTS WERE EVIDENCED. WHEN POSITIVE AND NEGATIVE EXAMPLES WERE GIVEN, CRITERION WAS REACHED RAPIDLY. GENERALLY, SUBJECTS ACHIEVED FEW CORRECT SOLUTIONS BEFORE REACHING CRITERION, AND BACKWARD AND FORWARD LEARNING CURVES SHOWED NO INCREASE IN PROPORTION OF CORRECT RESPONSE JUST BEFORE REACHING CRITERION. AN ALL-OR-NONE MODEL WAS REJECTED ON THE BASIS OF TESTS FOR INDEPENDENCE ON PAIRS OF ADJACENT RESPONSES AND FOR NUMBERS OF SUCCESSES IN BLOCKS OF THREE PROBLEMS. SUBJECTS LEARNED MEDIATING CONCEPTS, BUT REQUIRED ADDITIONAL SESSIONS TO APPLY CONCEPTS TO SPECIFIC PROBLEMS. (LB)