Kenneth Wachter asserts that the approximate relation that I proposed between the intrinsic growth rate of a population and that population's mean age-specific growth rate below age T, the mean length of generation, does not hold in general.' General reliance on the approximation is said to lead to error. I am surprised by this claim, because I was not proposing an exact, error-free relation between the two values but an approximate relation. Approximations are not to be judged by whether they are exact and error-free, but by how well they function as approximations. I demonstrated that the approximation worked well in simulations of the type of systematic changes in vital rates typical of the demographic transition, and in the type of abrupt decline to replacement-level fertility that demographers have studied under the rubric of population momentum. Errors were clearly present (and presented) in these applications, but they tended to be relatively small. The approximation is exact if the population is stable, if all childbearing is concentrated at one age, or if the population of childbearing age is stationary regardless of age-specific growth rates at other ages.2 The starting point for the approximation is the recognition that there must be some age in the childbearing interval in any closed population below which the mean of agespecific growth rates is exactly equal to the intrinsic growth rate. It is reasonable to look for this age in the middle of the childbearing period, i.e. around A*, the mean age of childbearing, or T, the mean length of generation. I argued that there was no a priori reason for choosing T, but that 'in most populations F(T) [the mean of age-specific growth rates below age T] will also be very close to the mean of age-specific growth rates below ages that are close to T, since the growth-rate function is normally subject to gentle changes with age and since F(a) is a mean of cumulated rates. There is no guarantee that F(T) will always, or even usually, perform better as a prediction of r [the intrinsic growth rate] than F(a) at some other age that could be specified a priori, although our simulations (not shown) indicate that it works better than F(A*) '.3