Purpose: The objective of the study was to compare test-retest reliability of three dichotic listening tests: SCAN-3 Competing Words Test (Words; Keith, 2009a, 2009b), Double Dichotic Digits Test (Digits; Musiek, 1983a), and Bergen Dichotic Listening Test With Consonant-Vowel Syllables (Syllables; Hugdahl & Hammar, 1997). Method: Sixty English-speaking children, 7-14 years old with normal hearing, had a single study visit during which each test was administered twice. Changes on retest were summarized by within-subject standard deviation ([S.sub.w]), compared among tests, and compared with binomial model predictions. Correlates of variance were explored. Results: Scores based on 40 items were more precise ([S.sub.w] = 5%) than those based on 20-30 items ([S.sub.w] = 6%-8%). All 3 tests had reliability within bounds predicted by binomial model. Changes on retest for Words and Digits Test were weakly associated with age, but this is confounded by the trend for older children to have higher Words and Digits scores. Conclusions: Digits Right, Digits Left, and Words Total scores--each based on 40 items--had the best reliability among the clinically used scores. Scores based on fewer items were less precise. Poor precision may contribute to misdiagnosis in clinic and to nondifferential misclassification in research. More precise estimates of dichotic listening ability require longer tests.