This paper presents an automatic subpixel registration algorithm that minimizes the mean square intensity difference between a reference and a test dataset, which can be either images (two-dimensional) or volumes (three-dimensional). The algorithm uses an explicit spline representation of the images and a coarse-to-fine iterative strategy (pyramid approach). The minimization is performed using a new variation (ML*) of the Marquardt-Levenberg algorithm for nonlinear least-square optimization. The geometric deformation model is a global three-dimensional (3-D) affine transformation, which can be restricted to rigid-body motion (rotation and translation) combined with isometric scaling. The method also includes an optional adjustment of image contrast differences. The algorithm is applied to intramodal positron emission tomography (PET) and functional magnetic resonance imaging (fMRI) data, demonstrating excellent results. The multiresolution refinement strategy is shown to be more robust than a single-stage method, reducing the likelihood of being trapped in false local optima. The improved Marquardt-Levenberg algorithm is also faster. The paper outlines the registration procedure, including the choice of data space, objective criterion, and transformations, and discusses the optimization, convergence, and practical issues. The use of cubic splines for interpolation and the benefits of a multiresolution pyramid are also detailed. Experimental results in an ideal case show the precision and speed of the proposed method.This paper presents an automatic subpixel registration algorithm that minimizes the mean square intensity difference between a reference and a test dataset, which can be either images (two-dimensional) or volumes (three-dimensional). The algorithm uses an explicit spline representation of the images and a coarse-to-fine iterative strategy (pyramid approach). The minimization is performed using a new variation (ML*) of the Marquardt-Levenberg algorithm for nonlinear least-square optimization. The geometric deformation model is a global three-dimensional (3-D) affine transformation, which can be restricted to rigid-body motion (rotation and translation) combined with isometric scaling. The method also includes an optional adjustment of image contrast differences. The algorithm is applied to intramodal positron emission tomography (PET) and functional magnetic resonance imaging (fMRI) data, demonstrating excellent results. The multiresolution refinement strategy is shown to be more robust than a single-stage method, reducing the likelihood of being trapped in false local optima. The improved Marquardt-Levenberg algorithm is also faster. The paper outlines the registration procedure, including the choice of data space, objective criterion, and transformations, and discusses the optimization, convergence, and practical issues. The use of cubic splines for interpolation and the benefits of a multiresolution pyramid are also detailed. Experimental results in an ideal case show the precision and speed of the proposed method.