The authors studied intra- and inter-reader reliability of the interpretation of cranial ultrasound examinations of very low-birthweight infants. A radiologist read 173 films, 88 of which he had read previously; the other 85 had been read previously by a second radiologist. For the diagnoses of subependymal hemorrhage and intraventricular hemorrhage, intra-reader agreement was similar to inter-reader agreement. 98 ultrasound films were read initially as showing subependymal hemorrhage; in nine cases the second reading did not agree. In five of 58 cases read initially as showing intraventricular hemorrhage, the second reading did not agree. Similarly, of 32 cases read initially as showing intraparenchymal echo-density, four were interpreted as negative on second reading. For all three diagnoses, disagreement occurred often enough to cause substantial misclassification bias when cranial ultrasound is used for clinical research.