The relationship between classical discrimination indices (CDI) and criterion-referenced discrimination indices (CRDI) and the appropriateness of each for use on criterion-referenced tests are investigated. A CRDI is proposed that attempts to separate those who master material from those who do not master material. A 26 item multiple-choice formative test (as well as a parallel true-false test) was given to 205 college undergraduates. The students were also given one period of instruction, and a table of specifications was made for their unit of learning. There were 32% "masters" (achieving 80% correct) and 68% "non-masters" within the sample on the multiple-choice test. The CDI and the CRDI were computed and compared with point biserial and tetrachoric inter-item correlations respectively. All 26 items were significant at the 0.05 level with respect to the CDI, whereas 22 items were significant with respect to the CRDI. There were 22 items in common that were significant for both the CDI and CRDI. It appears that classical indices tend to include more items than criterion-referenced indices. Futhermore, criterion-referenced items tend to be a subset of norm-referenced items with few exceptions. (Author/RL)