Autonomous manipulation is one the the most complex fields of humanoid robotics research, because it involves several domains such as: object recognition, object tracking, manipulator positioning, grasping, etc. In this paper a tactile controller based on fuzzy logic is proposed, capable of complex inhand manipulation tasks. This fuzzy logic approach assures finger motion smoothness by dealing with sensor nonlinearities and uncertainties, thus reducing the need of knowing the object dimensions or exact position in relation to the robot hand.