Fluid queueing models with finite capacity buffers are applied to analyze a wide range of real life systems. There are well established numerical procedures for the analysis of these queueing models when the load is lower or higher than the system capacity, but these numerical methods become unstable as the load gets close to the system capacity. One of the available numerical procedures is the additive decomposition method proposed by Nail Akar and his colleagues. The additive decomposition method is based on a separation of the eigenvalues of the characterizing matrix into the zero eigenvalue, the eigenvalues with positive real part and the eigenvalues with negative real part. The major problem of the method is that the number of zero eigenvalues increases by one at saturation. In this paper we present an extension of the additive decomposition method which remain numerically stable at saturation as well.