Variations of a certain kind, i.e. significant variations between whole populations. The number of different independent ways the theory predicts we should see evidence of these variations in present and past populations and the finding that they are all observed just as predicted, make it astronomically unlikely that they could be coincidental - and the fact that we have an observed mechanism that explains how the variations arise and has given us a precise mathematical model of how the variations spread within populations, means that it's arguably the best and most well-tested theory in science (pace quantum mechanics).
To deny it rationally, you need to come up with some good reason why the observed mechanism could not be the cause of those changes and why the various independent lines of evidence predicted by the theory are not evidence for the theory and come up with a better theory to explain all the independent lines of evidence.
Good luck with that.