New items are being calibrated for the next generation of the computerized adaptive (CAT) version of the Armed Services Vocational Aptitude Battery (ASVAB) (Forms 5 and 6). The requirements that the items be "good" three-parameter logistic (3-PL) model items and typically "like" items in the previous CAT-ASVAB tests have resulted in a set of algorithmic rules that are general enough to be of interest to researchers and others working to improve CAT tests. For CAT-ASVAB on-line calibration, a seeded-item scheme is being used. Answers to the seeded items (new) are collected along with answers to nonseeded items to allow the estimation of examinee ability and the evaluation of the quality of each new item relative to the ability scale. Six types of deficient items are discussed. The first five types lead to nonparametric item characteristic curves that cannot be fitted well with a 3-PL model. The sixth category, in which the item has insufficient information, is the most difficult to correct. A set of factors has been developed to help deal with these problems and increase the value of item information. (Contains 6 tables and 6 figures.) (SLD)