The effectiveness of a legged robot’s gait is highly dependent on the ground cover of the terrain the robot is traversing. It is therefore advantageous for a legged robot to adapt its behaviour to suit the environment. In order to achieve this, the robot must be able to detect and classify the type of ground cover it is traversing. We present a novel approach for ground cover classification that utilises position measurements of the leg servos to estimate the errors between commanded and actual positions of each joint. This approach gives direct insight into how the robot is interacting with the terrain. These position sensors are usually built into the actuators and therefore our approach has the advantage of not requiring any additional sensors. We employ a multi-class Support Vector Machine with a 660-dimensional feature space consisting of features in gait-phase and frequency domains. We implemented our algorithm in the Robot Operating System (ROS) framework for real time classification and also developed a MATLAB implementation for extensive offline testing. Both implementations perform multi-class ground cover classification with high accuracy across five classes.