A good criterion-referenced test item is defined as the one which allows the individual to answer correctly if he masters the criterion behavior represented by the item and answer incorrectly if he actually does not master it. Therefore, a good discriminating item for criterion-referenced tests is the one which has a larger proportion of correct responses in the mastery group and a smaller proportion of correct responses in the non-mastery group. Based on these considerations, the difference in proportions of correct responses in mastery and non-mastery groups and the phi coefficient are proposed as discrimination indices for criterion-referenced test items. These two indices were compared empirically with the point biserial correlation of items and test scores in three different situations: heterogeneous sample with a symmetrical distribution of scores, a homogeneous sample with a skewed distribution, and varying item difficulties. Results indicate the indices are highly correlated in most cases. Implications of these comparisons are noted. A possible approach for criterion-referenced test reliability is also discussed. (Author/DG)