Interobserver reliability in head circumference measurement was assessed in a cohort of 1105 low birthweight (less than or equal to 2000 g) infants enrolled in a study of brain hemorrhage. In 927 (83.9%) subjects, head circumference was measured both by a pediatric resident or admitting pediatrician, and by a trained ultrasound technologist. The Pearson correlation coefficient for these two sets of measurements was 0.934 (p less than 0.01), and the intraclass correlation coefficient was 0.933. By contrast, analysis of differences revealed that the limits of agreement were from -1.99 to 2.03 cm, indicating that 5% of measurements differed by 2 cm or more. Using clinicians' measurements as the gold standard, ultrasound technologists detected abnormal head circumferences with a sensitivity of 91.2%, a specificity of 97.2%, and a positive predictive value of 88.0%. For clinical purposes this level of reliability may be acceptable, but in research studies this degree of misclassification would lead to attenuation of the odds ratio.