To assess the impact of tax reform, it is necessary to take into account the way households adjust their taxable income to tax rates. The key parameter is the elasticity of taxable income to the marginal tax rate or its complementarity to one, a rate called "retention". This elasticity is estimated using individual data for the period 1997-2004. This period saw legislative changes to income tax (reduction of the ceiling of the family quotient and cuts in scale rates) which are "natural experiments" suitable for measuring the elasticity of taxable income with respect to the marginal rate, drawing on the estimation strategy used by Gruber and Saez (2002). Across all taxable households, the estimated elasticity of taxable income to the marginal retention rate is very low, in the order of 0.02. But there is a fairly strong heterogeneity of reactions, since it is 0.31 for the 10% of households with the highest taxable income. To illustrate, we examine the implications of such an elasticity in the context of a model to determine the maximum rate for the highest income, the Saez model (2001). This calculation is only indicative, however, because the Saez model is designed to assess the optimal rates on the assumption that income tax is the only redistributive instrument and the only source of revenue for public finances. It does not give clues as to the optimal rate to be applied in a tax system where income tax plays only a minor role.