This study is an attempt to answer the following research question: can the reliability of a criterion-referenced test be accurately determined according to a multiple classification of the student's performance? Specifically, the study pursues the beta-multinomial model, which postulates the probability distribution of an examinee's degree of mastery on a criterion-referenced test. From this model, a procedure for assessing the reliability of the testing instrument was developed. Simulated data based on the beta-multinomial distributions did not depart significantly from those generated by the beta-binomial model. However, these results should not preclude the utility of beta-multinomial models in this context. (Author/BW)