In this study, a robotic hand was equipped with force and bend sensors. Sensors were modified to fit the robotic hand and for more efficient utilization. A cylindrical grasping task was performed for three conditions, namely no object, soft object and hard object. Features were formed using the outputs of the sensors and their first and second derivatives. A multinomial logistic regression model was fitted to the data. Classification was done according to both object type (no object, soft object and hard object classes) and movement type (no movement, flexion, contact/release and extension classes). Results have shown that the information from the force sensors do not adequately contribute to the feature space because of poor coupling and this affects discrimination of soft object and contact/release classes. More sensors and a better actuation protocol need to be used in future work.