Virtual reality (VR) allows users to interact similarly to how they do in the physical world, such as touching, moving, and pointing at objects. To select objects at a distance, most VR techniques rely on casting a ray through one or two points located on the user’s body (e.g., on the head and a finger), and placing a cursor on that ray. However, previous studies show that such rays do not help users achieve optimal pointing accuracy nor correspond to how they would naturally point. We seek to find features, which would best describe natural pointing at distant targets. We collect motion data from seven locations on the hand, arm, and body, while participants point at 27 targets across a virtual room. We evaluate the features of pointing and analyse sets of those for predicting pointing targets. Our analysis shows an 87% classification accuracy between the 27 targets for the best feature set and a mean distance of 23.56 cm in predicting pointing targets across the room. The feature sets can inform the design of more natural and effective VR pointing techniques for distant object selection.