Managers are increasingly using computer models to predict prescribed fire behavior in various seasons and under different fuel conditions. Because fire behavior is related to fire effects on vegetation, validation of computer models will help understand possible outcomes and increase planning efficacy. Better predictions of fire behavior also will aid in management of risk associated with prescribed burning. We compared fire behavior predicted with BEHAVE, fire behavior and fuel modeling system, using standard and site-specific customized fuel models with observed fire behavior of strip headfires. These fires were observed in shortleaf pine (Pinus echinata)–dominated stands managed as pine–grassland stands for the endangered red-cockaded woodpecker (Picoides borealis). We evaluated the accuracy of fuel models across different seasons and fire return intervals.Fuels in all stands tended to be heterogeneous and discontinuous, with fuel loads differing considerably between growing and dormant seasons and time since burned both in weight and composition. Fuel models varied in accuracy depending on fuel loading and season of fire. Therefore, multiple fuel models were required to more accurately characterize fire behavior across fire seasons and fire return intervals. All fuel models failed to produce accurate predictions for fireline intensity and rate of spread. All fuel models tended to overpredict fireline intensity on low-intensity fires and underpredict fireline intensity on high-intensity fires. The firing pattern we chose may have influenced the accuracy of predictions, but we were unable to detect appreciable changes in wind speed due to presumed convective influences of backing and headfires approaching each other. Additional BEHAVE fuel models for a wider range of fuel conditions show some promise in providing managers with more realistic predictions of fire behavior under variable fuel conditions. However, field validation for specific sites is imperative.