This paper describes on-going work in the analysis of motion dynamics in pen-based interaction. The overall goal is the creation of a model of user motion in pen gestures where constraint and curvature vary over the length of the path. In particular, speed/curvature models of motion are used to analyze pen trajectories and infer target constraints obeyed by a user performing selection gestures. We aim to use this information to calculate an effective local spatial selection tolerance associated with each gesture. This can be used to perform selection according to user intent instead of their literal stroke. Here, we describe our early analysis of constrained user selection gestures, and outline a prototype application that infers a tolerance for one type of selection gesture. The application selectively splits pen strokes based on an analysis of user motion.