Data on control-informant agreement from four published case-control studies of Alzheimer's disease are compared, using both the kappa statistic and proportion of agreement for the presence and absence of exposures. Agreement was best for exposures involving lifestyle, medical interventions or disorders of more recent origin, and worst for exposures which involved judgements by the respondent. Agreement levels are similar across studies, and are commensurate with levels of specificity and sensitivity to be expected in this type of enquiry. We discuss the problems and implications associated with the interpretation of data from such studies of the elderly.