It is the most important sucess factor for the electricity generation industry to minimize operations cost of surplus electricity generation through accurate demand forecasts. Temperature forecast is a significant input variable, because power demand is mainly linked to the air temperature. This study estimates the information value of the temperature forecast by analyzing the relationship between electricity load and daily air temperature in Korea. Firstly, several characteristics was analyzed by using a population-weighted temperature index, which was transformed from the daily data of the maximum, minimum and mean temperature for the year of 2005 to 2007. A neural network-based load forecaster was derived on the basis of the temperature index. The neural network then was used to evaluate the performance of load forecasts for various types of temperature forecasts (i.e., persistence forecast and perfect forecast) as well as the actual forecast provided by KMA(Korea Meteorological Administration). Finally, the result of the sensitivity analysis indicates that a $0.1^{circ}C$ improvement in forecast accuracy is worth about $11 million per year.