This paper describes the development and focus in the analysis of the Brazilian metamodels used in the residential labeling system. The metamodels represent the thermal behavior of naturally ventilated and artificially air-conditioned residential buildings. A computational simulation experiment using the Energy-Plus program was conducted taking into account naturally ventilated and artificially air-conditioned residential buildings. Several parameters were considered for parametric simulations such as thermal transmittance, thermal solar absorptance, shading device, ventilation opening factor. The multi-linear regression method was used to predict the cooling degree hours and the annual energy consumption for heating and cooling for dormitories and living room. An assessment on the accuracy of the Brazilian meta models for residential buildings was analyzed and showed that the regressions include standard errors that may influence the metamodels accuracy. Therefore, artificial neural networks (ANNs) are presented as an alternative to improve the metamodels used in the Brazilian regulation. Comparing both statistical method multi -linear regression and artificial neural network method it can be seen that ANN presented lower results for standard deviation end root mean square errors. Therefore, ANNs are recommended to improve the accuracy of the prediction in further behavior associated with the labeling system. (C) 2015 Elsevier B.V. All rights reserved.