In this study, simulated and actual Air Force test data were used to compare the different procedures for equating mental tests: conventional (equipercentile and linear), Item Response Theory (IRT), and strong true-score theory (STST); data collection designs used were single-group, equivalent-groups, and anchor test. Equating transformations were evaluated by comparing equated scores with true/observed scores along with bias and root-mean-squared-error indices. Specific testing-condition manipulations included variations in test length, item difficulty, sample size, and examinee ability distributions. The study found that parallel subtests were best equated using the simple conventional methods; nonparallel subtests, on the other hand, were best equated with the more complex IRT and STST methods. There were few differences among the data collection designs when they were applied to samples of equivalent ability levels; the anchor-test design was essential for equating subtests using nonequivalent examinee groups. There was little advantage to be gained by increasing the sample size from 1,000 to 2,400 examinees. Equating accuracy was not markedly affected when subtest length was doubled, nor did it matter whether easy or difficult subtests were equated. (Author/BW)