In this paper, we focus on the explicit use of the spatial context of strokes in the recognition process of graphic gestures. We dispose of two sources of knowledge on the samples: their shape which is a classical one and their spatial context. The proposed method is based on three different points of view to exploit optimally these two sources of knowledge to perform the recognition. The first point of view uses the spatial context to filter the possible classes and then uses the shape to discriminate the remaining classes. The second one reverses the roles of the sources of knowledge. The third one uses them jointly. The underlying idea is that each point of view suits to one of the three case of source reliability. The challenge is to automatically compose the points of view and to combine them to build a system performing contextual shape recognition without any prior information on the targeted domain.