A method for solving filter networks made of linear and nonlinear filters is presented. The method is valid independently of the presence of delay-free paths in the network, provided that the nonlinearities in the system respect certain (weak) hypotheses verified by a wide class of real components: in particular, that the contribution to the output due to the memory of the nonlinear blocks can be extracted from each nonlinearity separately. The method translates into a general procedure for computing the filter network, hence it can serve as a testbed for offline testing of complex audio systems and as a starting point toward further code optimizations aimed at achieving real time.