Ball dribbling is a central element of basketball and a main challenge for creating basketball robots is to achieve stability of the periodic dribbling task. In this paper two control designs for ball dribbling with an industrial robot are compared. For the two strategies, the ball position is determined either through force/torque or visual sensor feedback and the ball trajectory is predicted with a recursive least squares algorithm. The end effector trajectory for each dribbling cycle is generated based on the predicted ball position/velocity at the dribbling height and the estimated coefficient of restitution. For both tracking approaches, dribbling for multiple cycles is achieved. The vision-based approach performs better as compared to the force/torque-based approach, in particular for imprecise estimates of the coefficient of restitution.