Artificial neural networks (ANNs) have been considered for assessing the potential of low GWP refrigerants in experimental setups. In this study, the capability of using R449A as a lower GWP replacement of R404A in different temperature levels of a supermarket refrigeration system is investigated through an ANN model trained using field measurements as input. The supermarket refrigeration was composed of two indirect expansion circuits operated at low and medium temperatures and external subcooling. The results predicted that R449A provides, on average, a higher 10% and 5% COP than R404A at low and medium temperatures, respectively. Moreover, the cooling capacity was almost similar with both refrigerants in both circuits. This study also revealed that the ANN model could be employed to accurately predict the energy performance of a commercial refrigeration system and provide a reasonable judgment about the capability of the alternative refrigerant to be retrofitted in the system. This is very important, especially when the measurement data comes from field measurements, in which values are obtained under variable operating conditions. Finally, the ANN results were used to compare the carbon footprint for both refrigerants. It was confirmed that this refrigerant replacement could reduce the emissions of supermarket refrigeration systems.