Alternate forms of the Armed Services Vocational Aptitude Battery (ASVAB) were administered five separate times to 57 men and women of military service age. The objective was to determine to what extent means and cross-session correlations are stable over several administrations. Ten individual subtests, the derived ASVAB area composites and the Armed Forces Qualification Test (AFQT) were examined for stability. The means for this sample were below the national average and scores were less dispersed. Means increased over sessions .5 standard deviation or more on half the subtests and consequently, on most of the composite scores. Correlations for the subtests and the composites were largely stable over sessions and were slightly higher later in practice. Reliabilities were comparable to reference populations when adjusted for the range restriction of the present sample. The implications of practice effects for paper and pencil as well as automated selection tests are discussed. (Author)