A maximum likelihood procedure is given for estimating household and community transmission parameters from observed influenza infection data. The estimator for the household transmission probability is an improvement over the classical secondary attack rate calculations because it factors out community-acquired infections from true secondary infections. The mathematical model used does not require the specification of infection onset times and, therefore, can be used with serologic data which detect asymptomatic infections. Infection data were derived by serology and virus isolation from the Tecumseh Respiratory Illness Study and the Seattle Flu Study for the years 1975-1979. Included were seasons of influenza B and influenza A subtypes H1N1 and H3N2. The transmission characteristics of influenza B and influenza A(H3N2) and A(H1N1) outbreaks during this period are compared. Influenza A(H1N1), A(H3N2) and influenza B are found to be in descending order both in terms of ease of spread in the household and intensity of the epidemic in the community. Children are found to be the main introducers of influenza into households. the degree of estimation error from the misclassification of infected and susceptible individuals is illustrated with a stochastic simulation model. This model simulates the expected number of detected infections at different levels of sensitivity and specificity for the serologic tests used. Other sources of estimation error, such as deviation from the model assumption of uniform community exposure and the possible presence of superspreaders, are also discussed.