When 74 sixth-grade special education students were tested on the fifth-and fourth-grade levels of the 1977 California Achievement Test, their scores dropped significantly, although converted to in-level norms. Test validity, measured by the number of students scoring at floor level, on the in-level versus below-level norms, was subtest dependent. There were no differences in the mathematics comprehension subtest, but significantly more students scored above floor level on the below-level reading vocabulary and reading comprehension subtests. A locator test, designed to aid teachers in selecting the proper test level, substantially underpredicted the percentage of students scoring above chance and floor levels. Results imply that out-of-level testing be approached with extreme caution. (CP)