A level set segmentation algorithm based on Hermite derivative filter is proposed for segmentation of human magnetic resonance images (MRI). Instead of utilizing the traditional first difference, Hermite derivative filter was used to calculate the differential coefficients in the course of level set interface evolution, so that the differential coefficients were no longer decided by the first neighbor, but by the second neighbor of the examined pixel. Results of the segmentation tests proved that to the same segmentation process, the level set method utilizing Hermite derivative filter produced a more accurate result. The proposed method showed special superiority over the conventional method for images with interferences by noise. At the same time, the new algorithm did not increase the time for the segmentation.