Instructional sensitivity indices are compared from four perspectives: (1) criterion-referenced testing; (2) classical test theory; (3) item response theory; and (4) Bayesian theory. Instructional sensitivity is defined as the tendency of a test item to vary in difficulty as a function of instruction. The instructional sensitivity of the items in the grammar subtest of the Michigan Test of English Language Proficiency was assessed as an example. Subjects (N=18) were enrolled full-time in proficiency level 3 at the Center for English as a Second Language at Southern Illinois University. The normalized difference between pre- and post-test logits of difficulty (item response theory) and the pre-to-post difference index (criterion referencing) were found to be the most robust. Numerous problems, including spuriously high indices, were encountered with the Bayesian indices. It is concluded that decisions about discarding or revising test items should not be based solely on instructional sensitivity indices. Five data tables are included. (TJH)