Emrick (1971) described an evaluation model for mastery testing that has since generated some interest. In this model it is assumed that 1) educational objectives consist of a collection of unitary and explicitly defined skills, 2) each of these skills is a binary (all or none) variable and 3) tests designed to assess mastery of a particular skill consist of collections of test items that are homogeneous in terms of content, form and difficulty. In addition, Emrick's model incorporates two types of error: Type I or alpha (a) in which the examinee's responses lead to a mastery conclusion when hir true status is nonmastery, and Type II or beta (/3) in which hir item responses lead to a nonmastery conclusion when in fact (s)he has mastered the skill in question. Table 1 was used by Emrick as a representation of the decision situation. Note that both a and A are conditional probabilities; thus, ( is the probability of a nonmastery decision given that the examinee has attained mastery of the skill in question. This implies that the display in Table 1 cannot be interpreted in the usual manner. Ordinarily the parameter (/, as shown in Table 1, would be interpreted as the probability that an examinee has attained mastery and gives an incorrect response, i.e., 0 would not represent a conditional probability. When the entries in a fourfold table represent the corresponding probabilities of four mutually exclusive events, the entries must sum to one. We see that the entries in Table 1 sum to two, not one. Moreover interpreting Table 1 in the usual way would imply that the number of masters and the number of nonmasters are equal, since the corresponding marginal proportions are both equal to one. We assume, therefore, that Emrick did not intend Table 1 to be interpreted as an ordinary fourfold contingency table, and more importantly, that a and , represent conditional probabilities, as indicated above. Let RR denote the ratio of regret of Type II to Type I decision errors (see Emrick, 1971, p. 324). The main result given by Emrick is an algorithm for determining the cut-off score for a mastery test, the algorithm being a function of a, /, RR, and the number of items on the test, say n. Our goal in this paper is to point out certain restrictions in Emrick's model that may not be immediately evident. As indicated above, the use of Emrick's model depends upon the values of a and /3. One difficulty is, of course, that the values of a and 0 are usually unknown. Consequently, it is necessary to devise a procedure for estimating these parameters. As a solution to this problem, Emrick uses a function of the average interitem reliability as an estimate of a + (. In particular, he attempts to estimate the correlation