An application of modulating function (MF) technique to the direct closed-loop identification for the estimation of a continuous-time model of multi-input-multi-output magnetic bearing system from sampled data are presented in this paper. The plant description given as a differential matrix polynomial is transformed into a matrix of difference equations using the modulating function approach, which allows to avoid numerical problems by aproximating time derivatives. Additionally, the initial conditions can be neglected as a characteristic of the MF technique. The system parameter estimation represents a solution of a least-squares problem derived from the difference system model with experimental data from an industrial magentic bearing system.