Nonparametric estimation of a discontinuity in regression

We propose and study a new method to nonparametrically estimate a discontinuity of a regression function. The optimal rate of convergence n - 1 is obtained under minimal assumptions. No smoothing is required.