We discuss the intercomparison between the Karlsruhe Optimized and Precise Radiative transfer Algorithm (KO-PRA) and the Reference Forward Model (RFM) codes, which have been designed for analysis of MIPAS-ENVISAT data. The purpose of this intercomparison is to validate the KOPRA algorithm, i.e. to identify and to remove possible errors in the KOPRA (or RFM) code and to quantify the reason of remaining differences. Similar comparisons between the MIPAS Optimized Forward Model (OFM) and the RFM as well as between KOPRA and the RFM have already been performed. To be able to relate on these results, this validation is similarly organized: we perform subsequently more complex tests of ray-tracing, integrated column amounts, homogeneous and limb path calculations of unapodised, apodised and field-of-view (FOV) convolved spectra, using the same isolated CO2 line as well as the same six MIPAS microwindows. Additionally we compare modeling of CO2 line-mixing, non-local thermodynamic equilibrium (NLTE), trace gas continua and cross-section spectra. The KOPRA-RFM residuals are below a quarter of the noise- equivalent spectral radiance (NESR) for the isolated CO2 line as well as for the MIPAS microwindows, i.e. KOPRA fulfills the acceptance criteria requested for the OFM. In most cases the deviations are even clearly below 1 nW/(cm2 sr cm-1), that is more than one order of magnitude below the acceptance threshold. This is valid for unconvolved as well as for ALS (apodised line shape) and FOV convolved spectra. There is also good agreement in modeling of the H2O-, O2- and N2-continua and of CO2 line-mixing. Larger deviations of up to several nW/(cm2 sr cm-1) occurred for NLTE calculations on the basis of 'default' atmospheric profiles with vertical resolution of 1 or 2.5 km. These differences were found to be due to different layer-averaging of the vibrational temperatures and could be considerably reduced by calculations with a higher vertical resolution of 250 m. Cross-section spectra agree well, if the tabulated data are given independent of pressure, e.g. for ClONO2 and N2O5, and cover the atmospheric temperatures. Due to different temperature extrapolation the deviations increase up to 10 nW/(cm2 sr cm-1) for atmospheric temperatures outside the measuring range. The RFM is not yet adjusted to cross-sections tabulated for non- equidistant temperatures and for atmospheric pressures, like CFC-data in the HITRAN96 database. If these data are used, larger differences arise, e.g. up to 30 nW/(cm2 sr cm-1) between CFC-12 spectra. Avoidance of interpolation by performing homogeneous path calculations for p,T of one of the tabulated cross-section datasets reduces the deviations to below 0.5 nW/(cm2 sr cm-1).