Some evidence suggests that lay persons are able to perceive sexual orientation from face stimuli above the chance level. A morphometric study of 390 heterosexual and homosexual Canadian people of both sexes reported that facial structure differed depending on the sexual orientation. Gay and heterosexual men differed on three metrics as the most robust multivariate predictors, and lesbian and heterosexual women differed on four metrics. A later study verified the perceptual validity of these multivariate predictors using artificial three-dimensional face models created by manipulating the key parameters. Nevertheless, there is evidence of important processing differences between the perception of real faces and the perception of artificial computer-generated faces. The present study which composed of two experiments tested the robustness of the previous findings and extended the research by experimentally manipulating the facial features in face models created from photographs of real people. Participants of the Experiment 1 achieved an overall accuracy (0.67) significantly above the chance level (0.50) in a binary hetero/homosexual judgement task, with some important differences between male and female judgements. On the other hand, results of the Experiment 2 showed that participants rated the apparent sexual orientation of series of face models created from natural photographs as a continuous linear function of the multivariate predictors. Theoretical implications are discussed.