It is challenging to directly measure the temperature distribution in solid oxide fuel cell (SOFC) due to their high operating temperature and the difficulty in placing temperature sensors. In this study, we have developed surrogate models to reconstruct the temperature distribution of a tubular SOFC with counter-flow arrangement based on the input of several local temperatures and their axial locations. The surrogate models were trained by the simulation results of 2500 cases. We have compared the accuracy and prediction time for the surrogate model with five different regression algorithms, which are respectively polynomial regression, random forests, classification and regression trees, neural networks, and support vector regression. We show that the surrogate model with a third-order polynomial regression algorithm delivers the best performance, which could reconstruct the temperature distribution of a tubular SOFC based on the information of six local temperatures in 1.4 s with a root mean square error of 0.027. Other regression algorithms examined in the present study are not appropriate for the SOFC temperature reconstruction due to the low accuracy of the prediction results. Their performance may be improved by properly adjusting the parameter settings. The present study helps understand the performance of the temperature reconstruction using different surrogate models, which could effectively simplify the temperature distribution measurement in SOFC stacks.