Thermal engineering related to industrial applications needs fast and accurate prediction for complex heat transfer problems. Proper orthogonal decomposition (POD) method can achieve this target. In this paper, the performances of POD interpolation and POD projection methods are compared for the purpose of method selection in those applications. The results show that under the same sampling condition, POD interpolation method can simulate the four-variable heat conduction problem quite accurately (maximum relative deviation is smaller than 1 x 10(-8)%) but losses accuracy for the six-variable heat conduction problem (maximum relative deviation is up to more than 50%); while POD projection method can obtain accurate results for both the problems (maximum relative deviations are only 0.187% and 1.83% respectively). Thus, POD projection method is more robust and adaptive than POD interpolation method for wider ranges of heat transfer problems. Moreover, the conclusion is also suitable for the guidance of applications in the fields of chemical engineering and fluid mechanics etc. Crown Copyright (C) 2012 Published by Elsevier Ltd. All rights reserved.