This article considers the problems of framing questions, the answers to which depend on the informants' memories. The problems are illustrated with examples from the British Survey of Sickness. In an experiment in which a group of government employees were asked to recall what sick leave they had taken and when, the predominant error consisted of forgetting when the leave occurred rather than forgetting it completely. The results of checks on other memory questions show the importance of carefully defining and redefining for the informant the period of time a question covers.