Recently, novel identification methods have been proposed based on orthogonal basis nonlinear functions. These methods present strong statistical convergence properties but have not been evaluated from a computational point of view. This paper investigates the computational cost and performance of an orthogonal basis nonlinear system identification method. The computational effort of the model estimation and simulation are evaluated for an increasing number of experimental data, regressor dimension and for different values of a bandwidth limiting parameter. Results show that complexity increases linearly with all parameters for first-order interactions. For second order interactions, complexity increases exponentially with the regressor dimension and linearly with the other parameters. A simulated example illustrates the obtained results.