Let's say someone was shown pictures of male and female faces, everyone with a middle-of-the-gender-road haircut, no makeup, same neutral expression on their face; that is, remove all gender-norm identifiers. If that person chose the males and females at a high rate of accuracy, that would mean that they are doing so not on the basis of gender norms, but on the basis of biological differences in the appearance of male and female faces.
What I think happens is that gender norms and biological sex differences both play a role in IDing, say, a stranger on the street.