Joint torques are an important parameter in the mechanical study of human movements. People’s mass properties and movement patterns have different effects on joint torques. As for human segments, measuring joint torques directly limits movement. Therefore, it is a more common practice to determine joint torques indirectly. Mathematical methods have been successfully used to indirectly determine joint torques. However, mathematical techniques can be challenging. Another way to identify joint torques is to use artificial intelligence techniques. In recent years, it has been seen that joint torque estimation algorithms based on artificial neural networks (ANN) give successful results. Electromyography (EMG) is widely used as input data in estimating joint torque with ANN. Obtaining EMG data is a difficult process and requires sensitive sensors. Data variability is an important factor for the joint torque estimation based on the ANN method. It has been seen that some studies have a limited number of participants with similar physical characteristics. In this study, the analysis of sit-to-stand movement was performed on 20 participants with different physical properties. Then, joint torques were calculated with the simulation model. After that, a four-layer neural network was trained using the angular displacements of the joints, segment heights, and segment mass as input data. Here, different ANN model variations were tested in terms of performance, and the best one was selected. It has been seen that the proposed ANN model shows high accuracy in estimating joint torques using a non-complex method.