Article dans une revue: We define generalized Pareto curves as the curve of inverted Pareto coefficients b(p), where b(p) is the ratio between average income above rank p and the p-th quantile Q(p) (i.e., b(p)=E[X|X>Q(p)]/Q(p))). We use them to characterize income distributions. We develop a method to flexibly recover a continuous distribution based on tabulated income data as is generally available from tax authorities, which produces smooth and realistic shapes of generalized Pareto curves. Using detailed tabulations from quasi-exhaustive tax data, we show the precision of our method. It gives better results than the most commonly used interpolation techniques for the top half of the distribution.