We present a new method for the atlas-free brain segmentation of proton-density-like 3D MRI images. We show how steerable filters can be efficiently combined with parametric spline surfaces to produce a fast and robust 3D brain segmentation algorithm. The novelty lies in the computation of brain edge maps through optimal steerable surface detectors which provide efficient energies for the rapid optimization of snakes. Our experimental results show the promising potential of the method for fast and accurate brain extraction.