The paper proposes and evaluates a strategy for the alignment of anatomical and functional data of the brain. The method takes as an input two different sets of images of a same patient: MR data and SPECT. It proceeds in four steps: first, it constructs two voxel models from the two image sets; next, it extracts from the two voxel models the surfaces of regions of interest; in the third step, the surfaces are interactively aligned by corresponding pairs; finally a unique volume model is constructed by selectively applying the geometrical transformations associated to the regions and weighting their contributions. The main advantages of this strategy are (i) that it can be applied retrospectively, (ii) that it is tri-dimensional, and (iii) that it is local. Its main disadvantage with regard to previously published methods it that it requires the extraction of surfaces. However, this step is often required for other stages of the multimodal analysis such as the visualization and therefore its cost can be accounted in the global cost of the process.