The Boston Naming Test (BNT) is one of the most commonly used tests of confrontation naming. The length of the test, particularly when administered to impaired patients, has prompted the derivation of several abbreviated forms. Short forms of the BNT have typically been equated in terms of difficulty, but not empirically derived for discriminating between normals and anomic patients. Furthermore, most reports to date have been limited in sample size and generalizability. The present study examined BNT data from a total of 1,044 subjects, including 719 normals and 325 patients with Alzheimer's disease (AD). Scores were calculated for the entire 60-item version as well as for eight previously reported short forms. The scores were examined for the effects of age, education, and gender, as well as for the ability of each form to discriminate between AD patients and normals. There was a significant effect of age, education, and gender on all previously published forms, and the short forms varied in their ability to discriminate between patients and controls. A stepwise discriminant analysis was conducted to empirically derive a new, gender-neutral short form with discriminability comparable to the full 60-item test. Norms from this sample on the empirically derived short form are reported.