To replace human activities by humanoid robots in hazardous environment, they should perform various motions, not only simply walking by their two legs but also passing through a narrow way, climbing a ladder and even getting in a operating machine. In such cases, humanoid robots are contacting/grasping the surrounding environment through their hands as well as their legs. In this paper, we propose a method to evaluate the postures of humanoid robots when their limbs (arms and legs) are constrained to the surrounding environment. From a given posture, a quantitative measure is calculated considering various aspects, such as joint torque limit, joint motion range, and the robustness from falling down and slipping down. Only static posture is evaluated, assuming stationary postures or quasi-static motions. Numerical examples show the validity of the proposed evaluation method. Based on the proposed evaluation, an example of optimizing a motion sequence is also shown.