Previous studies addressing the usefulness of pallor in anemia are deficient because of observer or spectrum bias. Three internists made individual assessments of conjunctivae, face, nails, palms, and palmar creases in 98 male and 5 female hospitalized patients at a Veterans Administration Medical Center. The true-positive rates (sensitivities) were highest for pallor at any one of three sites (ie, the conjunctivae, face, or palms; 0.65) and for pallor of the palms (0.53). True-negative rates (specificities) were best for palmar creases (1.00), for pallor at conjunctivae, face, and palms (all three in combination; 0.95), and for the face (0.90). Receiver operator characteristic curve analysis revealed that the examination of the nailbeds is inferior to all other sites or combinations. Interobserver variability K scores were negative for palmar creases but ranged from .16 to .51 for other sites. We conclude that (1) the absence of pallor does not rule out anemia, and therefore this sign is not useful for screening an asymptomatic population; (2) pallor of the conjunctivae, face, and palms together is of benefit in confirming the presence of anemia; and (3) neither the nailbeds nor palmar creases are of value in assessing the presence or absence of anemia.