Drawing curves in mid-air with fingers is a fundamental task with applications to 3D sketching, geometric modeling, handwriting recognition, and authentication. Mid-air curve input is most commonly accomplished through explicit user input; akin to click-and-drag, the user may use a hand posture (e.g. pinch) or a button-press on an instrumented controller to express the intention to start and stop sketching. In this paper, we present a novel approach to recognize the user's intention to draw or not to draw in a mid-air sketching task without the use of postures or controllers. For every new point recorded in the user's finger trajectory, the idea is to simply classify this point as either hover or stroke. Our work is motivated by a behavioral study that demonstrates the need for such an approach due to the lack of robustness and intuitiveness while using hand postures and instrumented devices. We captured sketch data from users using a haptics device and trained multiple binary classifiers using feature vectors based on the local geometric and motion profile of the trajectory. We present a systematic comparison of these classifiers and discuss the advantages of our approach to spatial curve input applications.