In this paper, we present an experimental study of a generic approach for developping on-line handwriting recognition systems. This approach allows building several types of handwriting recognition systems according to different needs arrised in pen-based interfaces. We evaluate different properties which give _exibility to our approach. We show that this approach may be used to build either writerindependent systems competing with current state-of-art systems on the UNIPEN database, or light single-writer systems able to learn with limited training samples.