An extensive set of covariances for neutron cross sections has been developed to provide initial, low-fidelity but consistent uncertainty data for nuclear criticality safety applications. The methodology for the determination of such covariances in fast neutron region is presented. It combines the nuclear reaction code EMPIRE, which calculates sensitivity to nuclear reaction model parameters and the Bayesian code KALMAN to propagate uncertainty of the model parameters onto cross sections. Taking into account the large scale of the project (219 ssion products), only partial reference to experimental data has been made. Therefore, the covariances are, to a large extent, derived from the perturbation of several critical model parameters selected through the sensitivity analysis. They define optical potential, level densities and pre-equilibrium emission. This exercise represents the first attempt to generate nuclear data covariances on such a scale.