In this era of sub-health, walking has been proclaimed as the best sport because it can gradually enhance body function and relieve stress. However, bad walking postures during sports time could cause unnoticed bad effect on the spine, joints, and muscles. It is necessary to develop a system to monitor the walking postures and remind people of the bad postures. There are some limitations of traditional gait monitoring systems, such as the scant number of walking postures to be detected and the specific using environment. This paper presents a walking postures estimation system with inertial measurement unit (IMU) that is able to detect the walking postures correctly, meanwhile the device of the system is easy to be attached. This system uses random forest method with the features using acceleration data and gyro date collected by IMU to classify walking postures such as bow legs and knock knees. According to the classification results, the presented system is able to classify different walking postures of users correctly.