A reduce order mode based inverse algorithm is developed in this paper. The reduced order model is developed with the POD-Galerkin technique. The inverse problem is resolved by the conjugate gradient method. The performance or the present inverse algorithm is examined by an inverse heat conduction problem of identifying the unknown time-dependent heat source in a complex region. The effects of the measurement point location and the measurement error on the performance of the inverse algorithm are studied thoroughly. It is shown that the present POD based inverse algorithm is very accurate as well as efficient even when the input data contain measurement error. A large time saving is also found by using the present inverse algorithm, it is more than 83 times faster than the CFD model based inverse algorithm.