In this paper we show that some of the predictions of models of consumer intertemporal optimization are in line with the patterns of nondurable expenditure observed in U.S. household-level data. We propose a flexible specification of preferences that allows multiple commodities and yields empirically tractable equations. We estimate preference parameters using the only U.S. micro data set with complete consumption information. We show that previous rejections can be explained by the simplifying assumptions made in previous studies. We also show that results obtained using good consumption or aggregate data can be misleading.