Twenty-four listeners received 20 dichotic listening runs of 30 pairs of natural CV nonsense syllables per run. Left- and right-ear responses are represented by two partially overlapping normal distributions with equal variance. Observed ear advantages across listening runs are also distributed normally. The means of the distributions of ear advantage are listener specific, but the standard deviations are 10.8%±2.5. The origin of reversals in direction of ear advantage among listening runs is statistical—a small mean/sigma ratio—and approximately six listening runs (180 listening trials) are required to achieve a split-half reliability coefficient of +0.90.