Recall food expenditure data, which is the basis of a great deal of empirical work, is believed to suffer from considerable measurement error. Diary records are believed to be more accurate. We study an unusual data set that collects recall and diary data from the same households and so allows a direct comparison of the two methods of data collection. The diary data imply measurement errors in recall food expenditure data that are substantial, and which do not have the properties of classical measurement error. However, we also present evidence that the diary measures are themselves imperfect.