This paper takes the form of advice given to a hypothetical government on the construction of a tax‐benefit model. The importance of the quality of the data and the care taken in its adjustment is emphasised. The choice of the unit of analysis, the requirements of the income data, methods of updating and the coverage and representativeness of the dataset as a whole is discussed. Caution in making use of estimates of behavioural change is advised. Throughout, examples drawn from the construction of TAXMOD, a model for the U.K., are used to illustrate particular points.