We analyzed the predictive accuracy of rate of fire spread models used operationally in Australia in five different fuel types (grasslands, temperate shrublands, semiarid shrublands, dry eucalypt [Eucalyptus spp.] forests, and conifer forests). This analysis was undertaken by comparing predictions of older models with those of newer models and noting changes in error statistics based on independent evaluation of datasets composed largely of wildfire observations. We observed the newer models to have improved prediction accuracy in four of the five cases over their previous counterparts; only in the case of the semiarid shrublands was there essentially no difference. Mean absolute errors were reduced between 56 and 70 percent. This study has highlighted the value of continuous improvement in fire behavior modeling.