An original algorithm has been developed which applies several new concepts to the problem of local models. The algorithm is fully simulator independent and user-transparent. It uses a sequential least squares procedure to build appropriate approximating formulae from a general model that contains ah the terms necessary to represent any particular mixture. No a priori information on the type of mixture (ideal/non ideal, narrow/wide boiling,...) or operating conditions of the process is necessary. Though, it is found that sophisticated simulators and integrators can suffer from the discontinuities introduced when parameters are updated or the local model is changed. In this regard, a regressive least squares method would lead to better results. Finally, local models algorithms are proposed as a new tool for the interpolation of properties in raw experimental tables.