The kappamax reliability index of domain-referenced tests is defined as the upper bound of kappa when all possibile cutoff scores are considered. Computational procedures for kappamax are described, as well as its approximation for long tests, based on Kuder-Richardson formula 21. The sampling error of kappamax, and the effects of test length and of score variability on kappamax are discussed. Estimates of kappamax for double-length tests are shown to be related to Spearman-Brown estimates. (CTM)