[eng] This article proposes a methodology for the construction and simultaneous evaluation of black box models for screw type water chillers. The estimation method used was generalized Least Squares. The proposed methodology is an iterative process and its evaluation is based on quality parameters of goodness of fit and statistical assumptions. Ten mathematical models with different levels of complexity were proposed. The statistical software used was Eviews 7.0. The selection of the models was based on the Occam's razor, remaining as the most feasible model is model 5 characterized by a multiple linear regression