Prompt Fission Neutron Spectra (PFNS) are very important nuclear data for reactor neutronic calculation tools. Most of the international evaluated nuclear data libraries lie on the Madland-Nix model, which is a based on evaporation theory of fission fragments. But very scarce data can be found regarding the PFNS covariance matrix associated to these evaluations. As an illustration of the impact of the PFNS on neutronic calculations, we will show a Monte-Carlo calculation of the neutron flux received by a PWR vessel, using different PFNS evaluations. The neutrons that have the highest probability to contribute to the vessel flux are those that are emitted at the highest energies; however most of the fission neutrons are emitted around 2 MeV. These results show the necessity to have very precise PFNS evaluations, and a proper estimation of associated covariances. The estimation of the PFNS covariance matrix associated to a model, after adjustment of model parameters, will be shown. This is performed by the CONRAD code, developed at CEA Cadarache. The final goal of the study is to adjust the parameters involved in fission fragments de-excitation in the FIFRELIN Monte-Carlo code, also developed at CEA Cadarache, which computes the PFNS among other fission quantities, and to provide the associated PFNS covariance matrix. However for the moment we focused the study on three historically widely used PFNS models: Maxwellian, Watt and Madland-Nix models. The covariance matrix on the adjusted spectrum comes mainly from the systematic uncertainty on some experimental parameters — namely the normalization, background, detection efficiency, etc. In order to propagate this type of uncertainties properly, the marginalization technique is used. A close knowledge of the conditions in which a particular experimental PFNS has been measured is required, in order to have a correct estimation of the PFNS uncertainties after adjustment. In this work, we propagated the uncertainty on normalization of experimental spectra, and the uncertainty on the energy-dependent neutron detection efficiency. We show the resulting PFNS and associated covariance matrix in the case of thermal neutron-induced fission of 235U and 239Pu.