Abstract. The aim of this paper is to show a practical experience in parallelizing a large CFD code starting from a sequential version. Major steps developing this activity for a 30000-Fortran-statements code are showed. This is done focusing on difficulties, techniques, "open" problems and solutions that a multidisciplinary group faces during the work. As a case study, the parallel implementation of a 3D Thin Layer NavierStokes code, originally developed in collaboration with Alenia and NLR , for the simulation of the subsonic/transonic flow around a complex aerodynamic configuration is considered. The message passing, for the data transmission among processors, is handled by PVM. The target parallel architectures are several MIMD machines (including cluster of workstations); a comparison of the results, in terms of performance, with the current most powerful supercomputers is presented.