Processing noisy point data is a tedious task. It often requires smoothing and meshing, followed by mesh processing algorithms. In this work we present a robust filter which operates directly on points using higher order surface approximations, computed iteratively. We extend the bilateral filter by using and filtering surface curvature, and introducing a quality parameter based on our local surface approximation. Using this process input point data is denoised and smoothed, while preserving features. We also extend our filter to allow feature enhancement, by controlling shock formation resulting in sharp edges in regions of high curvature. We show results of our algorithms on several scanned point sets.