Fire behavior predictions and forecasts are vital to tactical planning on wildland firefighting incidents. One major source of uncertainty in fire behavior predictions is spatial variation in the wind fields used in the fire models. In most cases wind data are limited to only a few specific locations, none of which may be actually near the fire location. Mountainsides, valleys, ridges, atmospheric stability, and the fire itself, influence both the speed and direction of wind flows. Fire analysts, meteorologists and fire managers are left to general forecast information, guesswork and expert opinion for estimating spatial wind variability. These methods provide rough estimates of wind speed and direction at ridges and to a lesser extent in the valleys, but are subject to local knowledge and the skill level of the analyst. Short-range meteorological forecasts and fire behavior calculations (e.g. 6-12 hrs) on large fires could greatly benefit from information on local winds over the entire terrain, at resolutions of 10 to 100 meters. The study described herein has three objectives: 1) develop a methodology for using commercially available computational fluid dynamics (CFD) software to produce high resolution surface wind maps and 2) quantify the effect of high resolution surface wind data on fire behavior predictions, and 3) address the fundamental science question of the practical potential for modeling fire-induced changes to the wind fields. This paper describes preliminary results of the modeling process applied to fire incidents in the Northern Rockies during the 2003 fire season.