The study compared two promising item response theory (IRT) item-selection methods, optimal and content-optimal, with two non-IRT item selection methods, random and classical, for use in fixed-length certification exams. The four methods were used to construct 20-item exams from a pool of approximately 250 items taken from a 1985 certification exam in a health field. Mastery status on the criterion test was determined for candidates by administering the full item pool. For the two IRT methods, separate test versions represented cutoff scores of 65%, 70%, and 75%. Results compared: (1) overlap in items selected; (2) exam information curves; (3) accuracy of decisions resulting from use of the exams. Optimal exams typically provided three to four times more information near the cutoff scores than exams constructed by the random method. The content-optimal method performed nearly as well as the optimal method. These results support the possibility of shortening conventionally constructed credentialing exams without losing decision accuracy. For a pre-specified exam length, optimal item selection can improve decision accuracy over non-optimal methods. (LPG)