Based on variable structure system theory and sliding mode, we develop an identification scheme suitable for time-varying parameters. The new identification scheme, working in closed-loop, addresses several key issues in system identification simultaneously: unstable process, highly nonlinear and uncertain dynamics, fast time varying parameters and rational nonlinear in the parametric space. Other important issues associated with identification, such as the persistent excitation property and insensitivity to measurement noise, are also discussed. A robotic manipulator is used as the illustrative example and the basis for comparison. (C) 2003 Elsevier Science Ltd. All rights reserved.