A model of the decline in frequency of an unnecessary virulence gene in a pathogen population following the mass removal of cultivars with the corresponding resistance gene in an agricultural situation is presented. The model is based on both forward and backward mutation and on selection for virulence and for avirulence in the presence and absence, respectively, of the allele for resistance in the host population. The model is used to estimate the fitness difference between avirulent and virulent races from field data that show declining virulence gene frequencies with time. One set of data was the decline of virulence in Puccinia graminis tritici (the cause of stem rust) to the Sr6 resistance gene in wheat (Triticum aestivum) in Australia between 1948 and 1955. The second was the decline of virulence in Erysiphe graminis hordei (the cause of powdery mildew) to the Mla6 resistance gene in barley (Hordeum vulgare) in the UK in 1969-1975. In both cases the estimated fitness differences were of the order 4-6%, the avirulent races being fitter. These values are considerably smaller than has frequently been assumed and imply that the use of multiline cultivars is more likely to provide directional selection in favor of complex rather than simple races of pathogens.