In this paper, we present a method that focuses on posture correction for stable quadruped locomotion over un- even terrain. Stability is ensured by switching to stable postures during gait transitions, where the posture is selected based on the terrain, foothold reachability and gait sequence. For fast and efficient posture evaluation, we use value functions that approximate stability and kinematic parameters. Learning using regression methods is used to create the value functions, which eliminates the need for additional sensors and computation for posture evaluation. This approach has been verified both numerically and experimentally. © 2017 ACM.