Abstract : The rise of pen-enabled interfaces is supported by the development of automatic methods for interpretation of more and more rich and complex input data: handwritten text, mathematical equations, sketches, free handwritten notes... For efficiently recognizing these handwritten documents, one has to consider jointly the shapes of their components and the relative positioning between them. Our research focuses on the modeling of relative positioning between handwritten objects, considering that the potential of this part of the information is not fully exploited in the current methods. We introduce spatial meta-templates, a generic modeling for describing spatial relations between objects of diverse nature, complexity, and shape. These models can be trained from data and provide richer and more accurate descriptions because they authorize to reason about spatial information directly in the image space. Relying on fuzzy sets theory and mathematical morphology allows dealing with imprecision and offers intuitive description of spatial relations. A meta-template is endowed with a prediction capacity: it provides the description of modeled spatial relations with respect to a reference object in the image, as a spatial template. This enables to conduct segmentation of objects depending on their spatial context. By exploiting these models, we present a new representation for structured handwritten objects. It relies only on modeling of the spatial information so as to demonstrate the importance of spatial information for interpretation of these objects. The segmentation of handwritten strokes into structural primitives is driven by positioning models, making use of their prediction ability. Experimental results, obtained with objects of diverse nature and complexity (Chinese characters, editing gestures, mathematical symbols, letters), validate the quality of positioning description offered by our models. The performance on the task of recognizing symbols with a spatial-based representation further attests the importance of this information and confirms the ability of meta-templates to model it properly and accurately. These results both show the richness of spatial information and give an insight on the potential of meta-templates for improving methods for handwritten document interpretation.