There are two problems in conventional linear regression model for forecasting monthly electricity sales that affect precision:the model ignores the fact that there are no heating measures and cooling measures in comfortable temperature range; the model also ignores the influence of random factors on monthly electricity sales because of random factors are difficult to quantify. To solve the above two problems, this paper puts forward two improvement measures:selecting low threshold temperature and high threshold temperature, and producing heating or cooling measures only when the actual temperature is below low threshold temperature or above high threshold temperature; proposing a method to quantify factors and putting the quantization value into prediction model as a factor of monthly electricity sales. The improved model can not only establish the relationship between temperature and monthly electricity sales preferably but also can account for the effect of random factors on monthly electricity sales reasonably, so the proposed measures are useful to improve precision. Making simulation analysis with actual data of Tongliang district in Chongqing, and the results show that the two kinds of improvement measures are effective.