A manual human machine interface operation system and method thereof are disclosed. In embodiment, this manual human machine interface operation system extracts user's arm image and palm image from the images capture by at least two cameras, and then calculates user's arm coordinate, so that user can select the object shown on the human machine interface by manually using his/her arm. The system then recognizes a hand posture according to the palm image and determines an instruction mapped to this hand posture. This instruction is then performed on the selected object. In embodiment, the hand posture can be a pointing posture, grab posture or a release posture.