FES has been used for restoring drop foot after the occurrence of an upper motor neuron disease, such as a spinal cord injury or a stroke. For the correction of drop foot in normal level walking, the common peroneal nerve is electrically stimulated by the control of a heel sensor installed in the sole, during swing phase of the gait. In this study, instead of the heel sensor, we developed a detection system to estimate swing phase of the gait by combining neural network learning FES with an inertial sensor. A surface electrode for FES and an inertial sensor are fixed on the tibial tubercle of knee, and the developed system is the compact system that a sensor part and an electrode part don't separate. The purpose of this study is to evaluate the performance of detection system of gait cycle not only normal level walking but also stepping stairs including the landing.