The goal of this paper is to design an online control interface for knee prosthesis based on the electromyography (EMG) signals of active thigh muscles. According to the time dependent nature of electromyography signals, translating such signals into precise commands in practical applications is a challenge for scientists. First stage for designing an online control interface is to design and implement a test setup for examining the proposed online control interface. To serve this purpose, active knee prosthesis is designed and manufactured using an elastic actuator mechanism. In order to measure the EMG signals, active muscles were detected based on the fundamental of muscles anatomy. In the second stage, filtering and data segmentation were utilized for electromyography signals smoothing, decreasing noises and reducing signal dimensions. Furthermore, time-delay neural network was used in order to map time domain features of EMG signals onto kinematic variables of knee joint. The angle and angular velocity of knee joint were estimated with accuracy of 0.85 (R²) for two locomotion modes including non-weight bearing and ground level walking. To implement online estimation of angular position, time domain features and neural network with 50 hidden layer’s neurons and 2 seconds time delay were used. Finally, online angular position estimation of knee joint was implemented on the designed test setup and results confirm proper tracking of online control interface.