Natural gas Local Distribution Companies (LDCs) need to estimate their customers’ gas demand accurately. A significant factor in the process of forecasting gas demand is historical data. This article presents a method of detrending historical data using linear regression and mathematical modeling applied to natural gas consumption. We present a detailed explanation of detrending the historical data with an annual two‐parameter model and a five‐parameter model at higher frequency. Our goal is to make all historical data look like it occurred during the most recent heating season. By using this method, demands from heating seasons before the most recent can be adjusted by adding demand proportional to the difference in the model terms so that historical daily data for different years become approximately stationary. The benefit of this detrending is demonstrated with an example of building forecast models with and without detrending on different length training sets.