Wildland fire managers will soon have access to computer models that simulate high resolution weather and fire growth for operational planning. Until recently, the use of these models has been restricted to research, primarily because of the substantial computing requirements of the high resolution weather model. The steady advances in computing technology have made weather modeling applications much more accessible. However, a case study of an integrated weather/fire spread simulation shows that predicting fire growth in complex terrain remains a formidable challenge. This paper compares the simulated and actual growth of the Bee Fire, which occurred in the San Bernardino National Forest, California, in summer 1996. Weather conditions at the time of the fire were simulated with a mesoscale model which provided a 2 km grid spacing. Fire growth was simulated with the FARSITE fire modeling system. A new statistical methodology was devised to evaluate the model performance, construct error bounds on projected perimeters, and statistically correct model biases.