Experimentally initiated epidemics of infectious pancreatic necrosis in rainbow-trout fry were analyzed using a modification of the standard mathematical model for a simple propagative epidemic. Contrary to expectations, the value of the transmission parameter (beta) was inversely related to initial density of susceptible hosts. This anomaly can be explained if we assume that the experimental epidemics were point-source rather than propagative epidemics. The implications of this conclusion for modeling experimental and natural epidemics are discussed.