Administrative data on personal income taxes and household budget surveys differ in at least one important respect: the definition of the unit of observation. The sociological concept of a household does not coincide with the administrative definition of a fiscal unit. We investigate whether the evaluation of a reform of income taxes is sensitive to this difference. The empirical results are obtained for a major reform of personal income taxes in Belgium in 1988. We use the technique of statistical matching to link the fiscal data of the Ministry of Finance with the household budget survey. We find that the characteristics of the tax system before and after the reform, such as liability progression and residual progression, are sensitive to the unit of observation and to the data set used. But this sensitivity evaporates at the level of the reform.