A major assumption of the biometric analysis of infant mortality as developed by Bourgeois-Pichat is that the age structure of infant deaths after the first month of life is virtually constant across time and cultures. Reanalysis of results from studies which compare the mortality of infants according to the type of feeding indicated that the relationship between mortality and age within the first year of life followed different patterns for breast fed and artificially fed infants. Historical data for populations with different breast feeding customs reveal similar differences in the age pattern of infant mortality. In populations where breast feeding was uncommon or of very short duration, infant mortality rises particularly steeply during the early months of the first year of life. The age structure of infant mortality in less developed countries where breast feeding is decreasing rapidly may be similarly affected. When substantial deviations from the linear relationship are evident, particular caution is required in applying the biometric technique, since in such situations the estimated endogenous mortality is very much affected by the particular set of data points within the first year of life which are chosen for the basis of the estimates.