It has been argued that visual subliminal processing of face-inherent sex information might depend on the retino-collicular projection. In the current study, we tested this possibility. Based on the known exclusive sensitivity of the retino-collicular projection for low spatial frequencies (LSF), we derived the following predictions. First, masked and, therefore, subliminal high-spatial frequency (HSF) face primes would not be processed, but, secondly, masked unfiltered face primes, thirdly, masked LSF face primes, and, fourthly, unmasked HSF primes should be processed and should lead to a congruence effect—that is, faster responses (better performance) in congruent conditions, with face primes and face targets of the same sex, as compared to incongruent conditions, with face primes and face targets of different sexes. These predictions were borne out in altogether three experiments. In Experiment 1, we observed that masked unfiltered primes created a congruence effect and that masked HSF primes failed to do so. In Experiment 2, we found that masked unfiltered primes and masked LSF primes both created significant congruence effects of about the same size. Finally, in Experiment 3, we demonstrated that unmasked HSF primes led to a congruence effect, whereas masked HSF primes failed to do so. Also, masking led to the subliminal presentation of the primes, as indicated by the fact that our participants were unaware of the masked but not the unmasked face primes. Together, these findings are in general agreement with an assumed origin of subliminal (or unaware) face processing along the magno-cellular projection of the human visual system.