System identification has long been of prime interest in industrial applications. The purpose of the current contribution is to present experimental on-line identification of a three-mass electromechanical system in digital input-output model. Studies are carried out by formulating the mathematical model of the plant using differential equations, and discrete-time identification using on-line plant input-output data. The theory and computer-based parameter identification method is described. Steady-state and transient behaviour of the system is investigated to define characteristics of the system. Recursive least squares method is used to estimate the unknown parameters of the system that match input-output behaviour of the real electromechanical system. Discrete-time data for parameter identification are obtained experimentally carrying out on a PM-DC motor set-up. Direct identification method is used in closed-loop identification. Root-mean-square error criterion is used for model validation. Results obtained in open-loop and closed-loop identification are presented which show variations in the system parameters. (C) 2003 Elsevier Ltd. All rights reserved.