Although below-knee prostheses have been commercially available for some time, today’s devices are completely passive, and consequently, their mechanical properties remain fixed with walking speed and terrain. To improve the current performance of below-knee prostheses, we study the feasibility of using the amputee’s residual limb EMG signals to control the ankle position of an active ankle-foot prosthesis. We propose two control schemes to predict the amputee’s intended ankle position: a neural network approach and a muscle model approach. We test these approaches using EMG data measured from an amputee for several target ankle movement patterns. We find that both controllers demonstrate the ability to predict desired ankle movement patterns qualitatively. In the current implementation, the biomimetic EMG-controller demonstrates a smoother and more natural movement pattern than that demonstrated by the neural network approach, suggesting that a biologically-motivated, model-based approach may offer certain advantages in the control of active ankle prostheses.