Thermal management systems of battery packs of electrical/hybrid electric vehicles play an important role in maintaining performance and longevity of battery cells. In order to optimize the thermal management system design, Computational Fluid Dynamics (CFD) is commonly used to accurately predict the thermal behavior of the system. However, the significant computational expense of CFD modeling means that its use for online predictive control and fast optimization design is infeasible. This paper presents the method of overcoming the obstacle via use of reduced-order models, which accurately approximate transient CFD solutions of conjugate heat transfer problems by solutions of strongly reduced (not more than a few tens of degrees of freedom) systems of differential equations. Algorithms for development of reduced-order models based on projection onto the Krylov subspace are presented for both single-input and multiple-input cases. The algorithms do not require access to the discretized equation matrix, which is usually inaccessible in a commercial CFD software. The reduced-order models are verified via comparison with full-order CFD simulations for a series of transient heat inputs. The comparison shows a reduction by a factor of about 1 0 3 in computational time without a significant loss of accuracy. • A drastic reduction in computation time is achieved. • The predictions of the ROM are practically identical to those of the CFD model. • The proposed method obviates the need for access to the discretized equation matrix.