We monitored prevalence ofNeozygites fresenii(Nowakowski) Batko in populations of cotton aphids,Aphis gossypiiGlover, during a 3–yr period and used these data to develop sampling strategies for predicting aphid population decline caused by disease. Aphid populations began to decline when average fungal prevalence (percentage of fungus–killed aphids) reached =15%. Populations declined to low levels 5–16 d later. Declines were more rapid in fields with higher initial densities of aphids. Fungus–killed aphids were detectable up to 10 d before prevalences reached =15%. Detection at this early stage required (theoretically) the examination of 70–150 leaves per field (95% probability of detection). However, a sample size of only 4–5 leaves was required for detection after prevalence reached 4%, which would be expected to occur =4 d before initial decline of aphid populations. With microscopic examination of 100 aphids per leaf, detection of fungus–infected aphids (95% probability) required between 9 and 17 leaf samples 10 d before decline of aphid populations, but only 1 sample was required 4 d before decline of aphid populations. During early stages of epizootics (prevalences<15%), fungal prevalences were greater on leaves 6 nodes below terminals, compared with leaves at the 1st or 2nd positions below terminals. During this time, fungal prevalences were correlated in subsamples collected 2 m apart. In samples collected 100 m apart during the very early stages of epizootics (prevalences<1%), distributions of infected aphids were clumped in 1 of 3 sample fields. These results underscore the importance of representative sampling of cotton fields if the goal is early prediction of epizootics.