In this work we address the problem of computing pointbased surface approximations from point clouds. Our approach is based on recently presented methods that define the approximated surface as the set of stationary points for an operator that projects points in the space onto the surface. We present a novel projection operator that differs from the defined in previous work in that it uses principal curvatures and directions approximation and an anisotropic diffusion equation to ensure an accurate approximation to the surface. We show how to estimate the principal curvatures and directions for point clouds and discuss the usefulness of the curvature information in the context of pointbased surface modeling and rendering.