The multiple linear models based piece-wise linearization approach is used to obtain control-oriented non-linear model of the non-linear multivariable wet grinding process of an industrial lead-zinc ore beneficiation plant. The overall process outputs are computed as a weighted sum of outputs from multiple models identified at several operating zones. The output weights of linear models are computed by a pseudo fuzzy supervisor. The linear models are identified using a transfer matrix identification technique using standard TF structures, and genetic algorithm based constrained optimisation formulation for the parameter estimation. A rigorous, plant validated model based wet grinding process simulator is used to obtain the identification and validation data for the proposed approach. Copyright © 2005 IFAC.