This study addresses the problem of parameters estimation for switched regression models used to represent systems with multiple operating modes or regimes. For the identification of such models, the collected data are from different operating modes and there is no a priori information holding on the partitioning of the data in regard to the different operating modes. The essential contributions of this paper lie first in the estimation procedure of the model parameters that provides an analytical solution, second in the simultaneous resolution of the problem of estimating the model parameters and allocating the data points to the different local models, and finally the robustness of the estimation procedure regarding the presence of outliers in the identification dataset.