Although the brain is not perfectly symmetrical with respect to the mid-sagittal plane, the automatic detection of this plane and of the degree of symmetry is of interest for many anatomical and functional studies. We propose a method for detecting the best symmetry plane in 3D MR brain images. We express this problem as a registration problem and compute a degree of similarity between the image and its reflection with respect to a plane. The best plane is then obtained by maximizing the similarity measure. This optimization is performed using the downhill simplex method and is initialized by a plane obtained from principal inertia axes, which proves to be close to the global optimum. This is demonstrated on several MR brain images. The proposed algorithm is then successfully tested on simulated and real 3D MR brain images. We also investigated the influence of the optimization procedure control parameters on the computation speed and result precision. Preliminary results obtained on CT and SPECT images suggest that the method can be extended to other modalities.