A method for training patterns validation for modelling MR damper is proposed. The method was validated with two models based on black-box and semi-phenomenological approaches. An input pattern that allows a better identification of the MR damper model was found. Including a frequency modulated displacement and increased clock period in the training pattern, the MR damper model fitting is improved. Also, the designed input pattern minimizes of training phase and reduces of the number of experiments. Additionally, incorporation of the electric current in the MR models outperforms the modelling approach.