Building energy modeling and simulation is an effective approach to evaluate building performance and energy system operations to achieve higher building energy efficiency. The high-order building models can offer exceptional simulation capacity and accuracy, however, its high level of complexity does not allow it to directly work with the optimization algorithms and methods that require a complete differential-algebraic-equations-based mathematical description of the physical model. In order to fill in the gap, the study presents a systematic approach to develop and calibrate the reduced-order building models. A notable feature of the approach is its coupling with high-order building simulations in order to pre-process the input information and support the calibration of the reduced model. A case study on a representative office building shows that the developed reduced-order model can present acceptable simulation accuracy compared with high-order simulations and significantly reduce the modeling complexity.