Multi-dimensional heat transfers modeling is crucial for building simulations of insulated buildings, which are widely used and have multi-dimensional heat transfers characteristics. For this work, state-model-reduction techniques were used to develop a reduced low-order model of multi-dimensional heat transfers. With hot box experiment of hollow block wall, heat flow relative errors between experiment and low-order model predication were less than 8% and the largest errors were less than 3%. Also, frequency responses of five typical walls, each with different thermal masses or insulation modes, the low-order model and the complete model showed that the low-order model results agree very well in the lower excitation frequency band with deviations appearing only at high frequency. Furthermore, low-order model was used on intersection thermal bridge of a floor slab and exterior wall. Results show that errors between the two models are very small. This low-order model could be coupled with most existing simulation software for different thermal mass envelope analyses to make up for differences between the multi-dimensional and one-dimensional models, simultaneously simplifying simulation calculations.