The permanent income hypothesis is tested on a four-quarter panel of about two thousand Japanese households for ten commodity groups. Consumption is a distributed lag function of expenditures, and the utility function is additively separable in time. Durability is defined as the persistence of the distributed lag. The permanent income hypothesis implies that, for each commodity group, expected change in expenditures is correlated neither with past expenditure changes on other commodities nor with expected change indisposable income, if its own lags are controlled for. The main results are the following: (1) durability is substantial even for food and services, (2)the permanent income hypothesis applies to almost all (probably more than ninety percent) of the population, and (3) the habit persistence hypothesis is rejected in favor of the permanent income hypothesis.