The stationary distribution for the population frequencies under an in
finite alleles model is described as a random sequence (x1, x2,...) su
ch that SIGMA x(i) = 1. Likelihood ratio theory is developed for rando
m samples drawn from such populations. As a result of the theory, it i
s shown that any parameter distinguishing an infinite alleles model wi
th selection from the neutral infinite alleles model cannot be consist
ently estimated based on gene frequencies at a single locus. Furthermo
re, the likelihood ratio (neutral versus selection) converges to a non
-trivial random variable under both hypotheses. This shows that if one
wishes to test a completely specified infinite alleles model with sel
ection against neutrality, the test will not obtain power 1 in the lim
it.