This paper presents an information fusion approach for automatic detection of mid-brain nuclei (caudate, putamen, globus pallidus, and thalamus) from MRI. The method is based on fusion of anatomical information, obtained from brain atlases and expert physicians, into MRI numerical information within a fuzzy framework, employed to model intrinsic uncertainty of problem. First step of this method is segmentation of brain tissues (gray matter, white matter, and cerebrospinal fluid). Physical landmarks such as inter-hemispheric plane alongside numerical information from segmentation step are then used to describe the nuclei. Each nucleus is defined according to a unique description according to physical landmarks and anatomical landmarks, most of which are the previously detected nuclei. Also, a detected nucleus in slice n serves as key landmark to detect same nucleus in slice n+1. These steps construct fuzzy decision maps. Overall decision is made after fusing all of decisions according to a fusion operator. This approach has been implemented to detect caudate, putamen, and thalamus from a sequence of axial T1-weighted brain MRI's. Our experience shows that final nuclei detection results are highly dependent upon primary tissue segmentation. The method is validated by comparing resultant nuclei volumes with those obtained using manual segmentation performed by expert physicians.