This paper proposes a new methodology for the estimation of ARMAX models, based on the implementation of a hybrid optimisation algorithm and a corresponding estimation procedure. The specific algorithm attempts to interconnect the diverse characteristics of two entirely different optimisation techniques, deterministic and stochastic, combining high convergence rate with increased reliability in the search for global optimum, and it consists of a super-positioned stochastic global search, followed by an independent deterministic procedure, in which the analytical gradients of the ARMAX model are used. The corresponding estimation procedure is split into two parts, due to the mixed linear-nonlinear relationship between the prediction errors and the parameter vector, and assures the stability and invertibility of the resulted models. The parametric identification test case, which is considered in this study, refers to the estimation of an ARMAX model for the description of a half-car passive suspension system of a road vehicle.