A virtual reality haptic system for capturing skillful human movements in control of a hemisphere rolling on a plane without slipping is presented in this paper. A dynamic model of this nonholonomic rolling system with configuration-dependent inertia and gravity is derived, and a solver, required for the real-time haptic interaction, is implemented. The performance of the haptic system is verified under experiments with human subjects. Experimental data recorded by the haptic system are analyzed and some common features of human movements in the precession phase of the manipulation of the rolling system are observed. Finally, a simple actuation scheme, capturing these features, is proposed and verified under simulation.