The paper presents an approach to manipulate unknown objects based on tactile information and force feedback. The object manipulation is performed using two fingers of the Shunck Dexterous Hand, which is equipped with tactile sensors on the fingertips. The contact point on each fingertip is modeled using a virtual articulation which adds a virtual degree of freedom to the finger. The approach uses the tactile data and hand kinematics information in order to estimate a grasp quality measure and to make finger adjustments after an initial grasp in order to improve the hand comfort. The approach was implemented in a real sensorized hand, and some examples manipulating different objects are presented in the paper showing the evolution of the resulting quality.