One of the larger sources of variation in human performance predictions is observer-to-observer variability. Whether because of differences in experience, training, motivation, innate ability, or perceived risk, observer performance varies. This paper explores observer variability in the task of detecting military targets in a FLIR image. Data for the analysis were obtained from a recent perception experiment conducted by NVESD utilizing 36 observers examining thermal imagery. Using the aggregate performance of the group as a measure of target difficulty, this paper compared the individual observer performance to that of the group. It was obvious that individual observers did not have identical performance characteristics. FOr example, targets detected by 50 percent of the group were detected by 20 percent of the observers over 70 percent of the time while another 20 percent of the observers detected then less than 20 percent of the time. Thus, the distribution of observer performance is fairly broad. This analysis of observer variability has implications for the modeling of target acquisition in combat simulations such as CASTFOREM and JANUS. These simulations currently use a uniformly-distributed, randomly-assigned threshold approach.