In order to develop a test of naming ability that is sensitive to changes in performance on repeated testing, but is unbiased by practice effects, the 85 items of the Boston Naming Test (BNT) were divided into two 42-item forms. Both forms were given to 15 healthy adult subjects, 24 patients with a clinical diagnosis of Alzheimer's disease, and 17 patients with other brain lesions. The reliability of the BNT was high as measured by coefficient alpha and interitem correlations. Performance on the two forms of the test was similar as indicated by mean scores and by correlations between scores. The BNT score includes uncued (spontaneous) and cued responses. When uncued responses were analyzed separately, the reliability and equivalence of the forms remained strong. Using uncued scores therefore does not alter the psychometric properties of the test, and considerably reduces administration time. The modification of the BNT described here should be useful in studies in which repeated measurements are obtained for the purpose of detecting changes in naming ability.