The dynamic relation between the displacement and reaction torque of the human joint is known as joint impedance. Properly quantifying joint impedance has medical potential in the diagnosis, understanding and modelling of movement disorders associated with neuromuscular conditions like cerebral palsy, stroke, dystonia and old age. The identification of joint impedance is often done with Linear Time Invariant (LTI) methods which lack the complexity to fully capture joint impedance over large operating ranges and over time. In this report a novel algorithm was developed which is able to identify joint impedance as a linear parameter varying system. This system description overcomes some of the limitations of the LTI methods. The algorithm was successfully tested in a simulation study in which it identifies a time-varying impedance model with a 5dB signal to noise ratio. Also, the developed method was applied on a force task with position perturbations done with the ankle and wrist. However, these data sets did not show sufficient time-varying behaviour and therefore the algorithm did not lead to better results compared to LTI methods. The reason the time-varying behaviour was not sufficiently excited was because of a faulty experimental protocol where the input was the main culprit.