Three approaches to the construction of achievement tests are compared: construct, operational, and empirical. The construct approach is based upon classical test theory and measures an abstract representation of the instructional objectives. The operational approach specifies instructional intent through instructional objectives, facet design, item forms, and item-generating algorithms. Criterion referenced and domain referenced tests are discussed, as they relate to the operational approach. A series of trials and subsequent calibrations are used in the empirical approach, which is based upon latent trait theory. To compare these three approaches, an achievement test was administered to 258 dental students, before and after instruction in basic dental anatomy. Five different strategies were used to construct the test forms: random selection, item-test correlations, pre-post test difficulty index, Bayesian, and Z-difference. Descriptive statistics were computed, including estimated and actual reliability, and estimated and true standard errors of measurement. The results suggested that a random sampling approach provides the greatest representativeness and minimizes errors of measurement. Random sampling is recommended for both the construct and operational definition approaches to measurement. A completely valid test of the third approach has not yet been completed. (Author/GDC)