This paper provides evidence on the relevance of the collection mode for the analysis of consumption data for the United States using complementary data sets from the Consumer Expenditure Surveys (CEX). We first show that population figures from consumption reports obtained with diaries markedly differ from those obtained using recall data. We then exploit multiple measurements of food expenditure available in the CEX to identify the effects of the collection mode on important features of the distribution of consumption (not just its mean). Finally, we show how to purge the expenditure measurements from most of the effects of the collection mode and thus obtain an improved measure of consumption that combines information from multiple reports in the CEX. The paper concludes by suggesting some guidelines for empirical research that have important implications for the measurement of inequality and well being.