The World Fire Web (WFW) is an image processing system developed by the European Commission, which contributes to better understanding the spatial-temporal distribution of global fire activity by detecting active fires. This system has been up-graded to process NOAA-AVHRR Pathfinder data in order to create long time series of active fire maps at the global scale for environmental studies. A first analysis of the original WFW products showed that the detection errors could vary from 95% in regions where the fire activity occur during the dry season to 5% between 60deg and 40deg latitude in the wet season. With the analysis of the spatial and temporal distribution of errors produced by the detection algorithm, this paper proposes and describes a new discriminant function which increases the detection accuracy of the system to greater than 95%