Abstract
The dominant model for income taxation in the public finance literature is the classical model of skills (Mirrlees, 1971). Until recently, an influential number of works using this model seemed to support declining marginal tax rates at high income levels. In this paper we use Diamond's (1996) methodology in order to explore the critical assumptions that lead to increasing or decreasing marginal tax rates. We find that with a lognormal distribution of skills and zero income effects there is a case for increasing marginal tax rates at high income levels. By performing a Kernel estimation to Israeli data we find empirical support for the lognormal distribution of skills.