We present here thorough experimental studies of a generic approach for developing on-line handwriting recognition systems. The basic principles of our approach lead to generic intrinsic properties that we investigate. These properties allow building various recognition engines corresponding to different needs in pen-based interaction.