I analyze the role of the distribution of skills in shaping optimal nonlinear income tax schedules. I use theoretical skill distributions as well as empirical skill distributions for 14 OECD countries. I find that a more dispersed log-normal skill distribution implies a more progressive optimal tax schedule. Optimal marginal tax rates should be lower throughout if a greater number of unskilled agents cluster at the bottom, and the scheme is more progressive if a greater number of agents locate at the top. I also highlight how the impact of the skill distribution is affected by the form of the social welfare function and the utility function. The findings using empirical skill distributions suggest that the results are sensitive to the type of statistical estimator used to estimate the skill distribution.