As noted by Satz, the prevalence of lateralized language in the general population is underestimated substantially by the proportion of people who show the typical asymmetry on a laterality task. In a series of two dichotic listening experiments with a total of 171 right-handers and 170 left-handers, we tested the hypothesis that increased reliability of measurement will lead to increased classification accuracy. Experiment 1 showed that neither the frequency nor magnitude of the right-ear advantage (REA) for fused rhyming words increased as the number of trials increased from 120 to 480. Ear-difference scores were highly reliable (r = .85), even when based on 120 trials. Experiment 2, which involved lists of dichotic word pairs, yielded similar results. Even though retest reliability of the ear-difference score for 132 word pairs was only .45, neither the incidence nor strength of the REA increased significantly when the number of pairs was increased to 528. The results indicate that the poor classification accuracy of dichotic listening tasks cannot be attributed to unreliability. (JINS, 2000, 6, 539–547.)