The invention discloses a temperature index construction method for a monthly power consumption prediction model. The method comprises the following steps of setting an air temperature threshold for distinguishing a power consumption load starting range, wherein a high-temperature threshold interval is [T-t, T+t], a high-temperature threshold is upt, a low-temperature threshold interval is [T-t, T+t], and a low-temperature threshold is dpt; performing grid searching in the high-temperature threshold interval and the low-temperature threshold interval according to a determined step length in order to determine all grid points, wherein the grid points are taken as a primary point set; obtaining a high-temperature index of an ith month according to the determined high-temperature threshold interval and all threshold grid points in the high-temperature interval, and obtaining a low-temperature index through similar processing; performing weighted synthesis according to weights of sub-regions within a whole region to obtain a whole-region air temperature index; and using all the high-temperature/low-temperature indexes formed according to the grid points as input data of the power consumption prediction model, predicting power consumption according to the power consumption prediction model, and obtaining a prediction error of the model. According to the method, a comprehensive index reflecting a whole-region air temperature is constructed, and a step-like influence of power consumption, an accumulative effect caused by high and low temperatures, and a regional difference reflecting an air temperature influence are depicted, so that the prediction ability of the model is enhanced.