The standard tests for infectivity of influenza viruses in eggs or mice have two major shortcomings. First, the host systems are known to be inhomogeneous, i.e. the response does not depend solely on the dose; and second, this variation from host to host cannot be assessed independently, since a single test only can be made on any one egg or mouse. Thus, the two probabilities—the presence of an infective unit in the inoculum and the success of a particular virus-host interaction—are confounded, so that we estimate not the number of infective units but an unknown function of this number. Valid comparisons of infectivity can still be made as long as one is satisfied with a relative answer, and does not wish to inquire into the nature of host-resistance.