A LEARNING MODEL TO IDENTIFY FACTORS CONTRIBUTING TO THE DIFFICULTY OF A PROBLEM ITEM WAS SUPPORTED EMPIRICALLY, AND INDICATED THAT THE NUMBER OF STEPS REQUIRED TO SOLVE A PROBLEM WAS THE MOST IMPORTANT VARIABLE IN PREDICTING BOTH ERROR PROBABILITY AND RESPONSE LATENCY. THE MODEL, IN ORDER TO ESTABLISH DIFFERENTIAL PREDICTIONS OF DIFFICULTY IN SOLVING ARITHMETIC PROBLEMS, IDENTIFIED SUCH VARIABLES AS THE MAGNITUDE OF THE LARGEST AND THE SMALLEST NUMBERS APPEARING, THE FORM OF THE EQUATION IN WHICH THE PROBLEMS ARE PRESENTED, AND THE DIFFERENCES BETWEEN NUMBERS IN SUBSTRACTIONS. ANOTHER VARIABLE, NUMBER OF STEPS (NSTEPS) REQUIRED TO SOLVE THE PROBLEMS, WAS FURTHER DIVIDED INTO TRANSFORMATION (STEPS REQUIRED TO PUT THE EQUATION INTO CANONICAL FORM), OPERATION (NUMBER OF OPERATIONS PERFORMED), AND MEMORY (NUMBER OF DIGITS THAT MUST BE HELD IN MEMORY). TERMINALS WERE INSTALLED IN EIGHT CLASSROOMS, AND 270 THIRD, FOURTH, FIFTH, AND SIXTH GRADERS PARTICIPATED FOR ONE ACADEMIC YEAR IN COMPUTER-ASSISTED MATHEMATICS INSTRUCTION. A REGRESSION ANALYSIS SHOWED NSTEPS TO BE THE MOST IMPORTANT VARIABLE IN PREDICTING BOTH ERROR PROBABILITY AND RESPONSE LATENCY, AND ANALYSIS OF THE FACTORS IN NSTEPS SHOWED THAT MEMORY WAS THE MOST IMPORTANT FACTOR. OPERATION PLAYED NO ROLE IN PREDICTING THE DEPENDENT VARIABLES. (OH)