We present a new symmetry-based method allowing to compute, reorient and recenter the mid-sagittal plane in anatomical and functional 3D images of the brain. In the literature, there are mainly two definitions of this plane: it is either the plane best fitting the inter-hemispheric fissure of the brain, or the plane best superposing the two sides of the head by reflective symmetry. We use this latter definition in our method, which is composed of two steps. At first, the computation of local similarity measures between the two sides of the brain allows to match homologous anatomical structures or functional areas, by way of a block matching procedure. The output is a set of point-to-point correspondences: the centers of homologous blocks. Subsequently, we define the mid-sagittal plane as the one superposing at best the points in one side of the head and their counterparts in the other side by reflective symmetry. The estimatio- n of the parameters characterizing the plane is performed by a least trimmed squares optimization scheme. Then, the estimated plane is aligned with the center of the image lattice. This method is fully automated, objective and reproducible. Our method tackles the main issue posed by the sym­me­try-ba- sed approach, that often relies on global similarity measures (such as the cross-correlation) between the intensities of the two flipped versions of the 3D image. The estimation of the mid-sagittal plane can be severely biased when normal or abnormal asymmetries hide the underlying symmetry of the brain or the skull. The computation of local measures of symmetry and the use of a robust estimation technique allow to discriminate between symmetrical and asymmetrical areas. Then, the mid-sagittal plane is mainly computed from the underlying gross symmetry of the brain, because its estimation is robust with respect to normal or abnormal asymmetries which are treated as outliers. We show on a large database of synthetic images that we can obtain a subvoxel accuracy in a CPU time of about 3 minutes, for strongly tilted heads, noisy and biased images. We present results on isotropic or anisotropic anatomical (MR, CT), and functional (SPECT and PET) images.