Heat flux measurement is important for dynamic thermal management (DTM) since it can provide more predicted information. Commercial heat flux sensors are too expensive and fragile for practical applications. Thermoelectric Cooler (TEC) based DTM provides the possibility of measuring the heat flux without additional heat flux sensor. Soft measurement of TEC heat flux has been presented via Neural Network (NN), which is suitable for identifying TEC model. NN models have been proposed by combing the classical TEC model together. The soft measurement performances with different NN models have been reported and analyzed in terms of accuracy, efficiency, and generalization capability.