In this paper, a dynamic identification method for a robot manipulator is investigated. A robust adaptive differentiator based on higher order sliding modes is used to estimate the state of the system. A theoretical proof of the convergence of this robust differentiator is given. A frequency analysis in term of filtering is provided and a comparison with a filter based on a classical differentiator is carried out. The dynamic parameters are estimated using the recursive least squares solution of a linear system obtained from the sampling of the dynamic model along a closed loop tracking trajectory. An experimental validation is performed on a robot manipulator and compared to another identification method.