Exoskeletons are successful at supporting human motion only when the necessary amount of power is provided at the right time. Exoskeleton control based on EMG signals can be utilized to command the required amount of support in real-time. To this end, one needs to map human muscle activity to the desired task-specific exoskeleton torques. In order to achieve such mapping, this paper analyzes two distinct methods to estimate the human-elbow-joint torque based on the related muscle activity. The first model is adopted from pneumatic artificial muscles (PAMs). The second model is based on a machine learning method known as Gaussian Process Regression (GPR). The performance of both approaches were assessed based on their ability to estimate the elbow-joint torque of two able-bodied subjects using EMG signals that were collected from biceps and triceps muscles. The experiments suggest that the GPR-based approach provides relatively more favorable predictions.