Since the primary purpose of classical testing is to rank order examinees consistently, the absolute value of the true score has been relatively unimportant. However, the major purpose of criterion referenced testing is to estimate the true capabilities of examinees to perform specific tasks. Hence, the problems of true score determination assume critical importance. Four measurement models which have potential for evaluating the results of criterion referenced tests are discussed here. The proportion correct model assumes that the proportion of sample trials scored correct is an unbiased estimate of the proportion correct in the infinite domain for that individual. The binomial error model adds the specification of the conditional distribution for the observed score for the given true proportion correct. The third model applies the philosophy implied in the binomial error model to Bayesian statistical theory, and the final model is the Rasch one parameter logistic model. Advantages and disadvantages of each are discussed, but the final choice of a model is to be based on the needs of the testing program and the resources available to analyze the data. (BW)