In this paper, it is proposed a new version of Normalized Least Mean Square (NLMS) algorithm, performed through a linguistic description based on the expert’s knowledge and implemented in a fuzzy rule base. In order to obtain a good trade-off between convergence speed and steady-state Mean Square Error (MSE) during the update of the estimate of the weights vector, it is used a Mamdani Fuzzy Inference System (MFIS) to adapt the step size of NLMS algorithm, whose inputs are the squared error and the normalized time instant. The validation of the proposed optimization algorithm was performed in Direct Adaptive Inverse Control (DAIC) design and applied to a non-minimum phase plant referring to an Electro-Hydraulic Actuator (EHA) system, shuch that a disturbance signal was added to the control signal.