This paper proposes a method t o identify t h e joint stiffness of a robot using a bandpass filter. It is based on moving one axis at a time. The dynamic model reduces to a model which is linear in relation to a minimum set of dynamical parameters which have to be identified. These parameters are estimated using the least squares solution of an over determined linear system obtained from the sampling of the dynamic model along a closed loop tracking trajectory. Conditions for a good data processing before identification are exhibited through practical aspects concerning data sampling and data filtering. An experimental study shows the efficiency of the method with t w o sets of data depending on motor and joint position measurements.