A major criticism of the Wechsler Memory Scale-III Faces subtest is the number of items, which can be daunting and time-consuming for an impaired client or boring for a normal client. An analysis of several versions, with data from a sample of 50 clinical referrals, revealed that a 32-item subtest was best overall. Using this version, 100% (Faces I) and 94% (Faces II) of the predicted raw scores were less than 4 points away from the actual raw scores, whereas 66% (Faces I) and 70% (Faces II) were less than 2 points away. Limitations of this procedure are discussed.