We propose a machine learning approach to the perception of a stable robotic grasp based on tactile feedback and hand kinematic data, which we call blind grasping. We ﬁrst discuss a method for simulating tactile feedback using a soft ﬁnger contact model in GraspIt!, which is a robotic grasping simulator . Using this simulation technique, we compute tactile contacts of thousands of grasps with a robotic hand using the Columbia Grasp Database . The tactile contacts along with the hand kinematic data are then input to a Support Vector Machine (SVM) which is trained to estimate the stability of a given grasp based on this tactile feedback and also the robotic hand kinematics. Experimental results indicate that the tactile feedback along with the hand kinematic data carry meaningful information for the prediction of the stability of a blind robotic grasp.