Family Resources Survey (FRS) data for April 1997 to March 2000 are used to estimate the take-up of income support (IS) by a subset of pensioners. We scrutinize the quality of FRS data for this purpose and describe a process of identifying and correcting inconsistencies in the data. Comparisons are made, before and after corrections to the data, of take-up estimates, logistic regression take-up models and predictions of take-up responses to changes in IS rates. Overall, the corrections do not have large effects on estimated take-up rates but suggest that non-take-up is marginally less serious than the uncorrected data imply. Logistic regressions using corrected and uncorrected data were in broad agreement on the factors influencing take-up. There were some differences in the scale of these influences, with implications for predictions of take-up responses to changes in the generosity of IS. Desirable improvements in the FRS are identified.