When criterion-referenced tests are used to assign examinees to states reflecting their performance level on a test, the better known methods for determining test length, which consider relationships among domain scores and errors of measurement, have their limitations. The purpose of this paper is to present a computer system named TESTLEN, which allows test developers to determine optimal criterion-referenced test lengths via simulation according to user-specified conditions. Such conditions may include ability distribution, item statistics, cut-off score, advancement score, test model, number of examinees, and number of replications. Output includes item statistics and values of decision consistency, kappa, and decision accuracy for each replication. The mean, range, and standard deviation of decision consistency, kappa, and decision accuracy are reported across replications. Also reported is the proportion of the examinee group assigned to each mastery classification. A school district developing a test as a diagnostic examination is used as an example of TESTLEN's practical application. Recommendations and directions for the use of TESTLEN appear in the appendix. (Author/AEF)