Methods for assessing the sibling relationship differ markedly, yet these have rarely been compared, or assessed in terms of retest reliability. We report an examination of sibling relationships as assessed by maternal interview, by videotaped observation in structured and semi-structured situations, and by naturalistic observation in unstructured settings. The different approaches to assessment were compared with data on 84 sibling pairs aged between 3 and 10 years. Test-retest reliabilities, assessed on 30 pairs studied on a second occasion, were good for maternal interview information, moderate for videotaped observations, and mixed for naturalistic observations. Independent positive and negative dimensions of the relationship were revealed by each method, and moderate agreement found between methods. Naturalistic observations of 30 minutes' duration were unsuitable for studying the negative aspects of the relationship.