Accurate energy demand forecasts are the main inputs for energy planners. However, to produce them close to actual demand values have always been a challenging task. Although traditional methods like multiple regression performed well in some cases, better performing techniques are still needed. Support vector regression is one of the state of the art technique developed in the last decade based on support vector machines. In this study, we have applied support vector regression to forecast monthly natural gas consumption of Istanbul. Our forecasting model outperformed traditional multiple regression and provide predicted values closer to actual consumption ones. Natural gas cannot be stored easily and can only be acquired through the purchase agreements. Accurate forecasts not only will reduce the natural gas purchasing costs of the city but will also diminish the risk of lack of gas substantially.