An anatomical landmark in the brain is a well defined point of the anatomy of the brain. Locating a landmark in a brain magnetic resonance image, or landmarking, consists of selecting a particular voxel in the image corresponding to the anatomical landmark in the imaged brain. This voxel, like an anchor, is a precious piece of information for performing measurements and registration of brain structures. Landmarking can be a tedious manual procedure, expensive and time consuming. It might be erroneous, difficult to assess, dependent on the scanner and on the landmarker. We have developed a generic algorithm that permits partial automation of the landmarking process. The algorithm has two components. One is an off-line procedure, the other is on-line. The former is a system that estimates the parameters of a probabilistic model from a training set of landmarked images using the Estimation Maximization (EM) algorithm. The latter inputs an image as well as the parameters previously estimated and outputs a tentative location for the landmark as well as a covariance metric that assesses the remaining uncertainty. The selected location can then be validated or corrected manually. An instantiation of the method for detecting the apex of the Head of the Hippocampus (HoH) is shown in the figure.