We present a method for calculating cost-of-living index numbers for arbitrary base period income levels without using heavy econometric estimation methods. The second order approximation formulas contain parameters which can easily be estimated by a differential demand system. Using a suitable specification, it is possible to work at a low level of commodity aggregation. The method is demonstrated on Netherlands data for the period 1952-1981. COST-OF-LIVING INDEX NUMBERS for individual households can be expected to differ substantially. There are various reasons for this expectation. Prices paid for the same commodity can vary between households. In principle each household has its own preference structure implying its own substitution behavior in the presence of changing prices. Income differences between households imply dif- ferences in consumption patterns which, in the presence of changing relative prices, imply diverging cost-of-living index numbers. Especially the variation of cost-of-living index numbers with the reference income levels attracts a lot of attention in the public discussion. There is some empirical evidence about the magnitude of this variation, because most statistical offices publish more than one consumer price index. The Netherlands CBS for instance has three CPI's, namely for all households, for households of wage earners with an income below a certain limit, and for households of wage earners above that limit. More detailed are the results of Michael (1979), Hagemann (1982), and Balk (1984). Using consumption patterns of individual households, obtained from a consumer expenditure survey, they compiled price index numbers for individual households and studied the differ- ences under various viewpoints. The drawback of these studies is that they are based on Laspeyres type price index numbers. Thus what in essence is compared are upper limits of household-specific cost-of-living index numbers. The substitu- tion behavior is ignored. This is acceptable only for relatively short periods or, more generally, for periods with relatively small price changes. useful feature of working with Laspeyres type index numbers is that one can use highly 1 The views expressed in this paper are those of the author and do not necessarily reflect the policies of the Netherlands Central Bureau of Statistics. The author gratefully acknowledges the computational assistance of H. A. van der Grient and A. J. Hundepool. Mike Baye is thanked for his comments on a previous version. previous version entitled A simple method for calculating