IN ORDER TO FIND A GOOD PREDICTOR OF EMPIRICAL DIFFICULTY, AN OPERATIONAL DEFINITION OF STEP SIZE, TEN PROGRAMER-JUDGES RATED CHANGE IN COMPLEXITY IN TWO VERSIONS OF A MATHEMATICS PROGRAM, AND THESE RATINGS WERE THEN COMPARED WITH MEASURES OF EMPIRICAL DIFFICULTY OBTAINED FROM STUDENT RESPONSE DATA. THE TWO VERSIONS, A 54 FRAME BOOKLET AND A 35 FRAME BOOKLET MADE BY DELETING UNNECESSARY FRAMES FROM THE FIRST BOOKLET, WERE TESTED IN THREE TEACHING MODES. ONE USED THE PROGRAM ONLY, ONE USED CONVENTIONAL TEACHING FOLLOWED BY THE PROGRAM, AND ONE USED THE PROGRAM FOLLOWED BY TEACHING. JUDGES RATED HALF OF EACH PROGRAM, FRAME BY FRAME, ON A TEN POINT SCALE OF FOUR CHARACTERISTICS--THE CONCEPT, THE VEHICLE, THE NUMERAL, AND THE RESPONSE. THE AVERAGE EMPIRICAL DIFFICULTY OF THE TWO VERSIONS DID NOT DIFFER, INDICATING THAT THE STEP SIZE AND NUMBER OF FRAMES DELETED ARE MOST LIKELY NOT IN ONE-TO-ONE CORRESPONDENCE. IT WAS CONCLUDED THAT JUDGES CAN RELIABLY ESTIMATE EMPIRICAL DIFFICULTY AND THAT THE RESPONSE AND NUMERICAL INDICES ARE MOST PROMISING FOR THIS PURPOSE. (BB)