"Research has shown that people are able to judge sexual orientation from faces with above-chance accuracy, but little is known about how these judgments are formed," says the abstract of a study by Joshua Tabak and Vivian Zayas posted today on PLoS ONE.

Science Codex jumps into the findings: "They cropped the grayscale photos so that only faces, not hairstyles, were visible. For women's faces, participants were 65 percent accurate in telling the difference between gay and straight faces when the photos flashed on a computer screen. Even when the faces were flipped upside down, participants were 61 percent accurate in telling the two apart. At 57 percent accuracy, they had a harder time differentiating gay men from straight men."

Going back to the full study—titled "The Roles of Featural and Configural Face Processing in Snap Judgments of Sexual Orientation"—Tabak and Zayas proceed with more detail:

[T]he finding that judgment accuracy remained above chance for upside-down faces strongly suggests that sexual orientation can be inferred from featural processing alone. Evidence suggests that if a trait can be inferred from featural processing alone, it may be inferred spontaneously and unintentionally in everyday life. Thus, the present results imply that in casual interactions, people may unwittingly accurately perceive others’ sexual orientation from brief glances at their faces. If so, it would appear that minority sexual orientation is not the concealed stigma that many argue it is. Indeed, the need to protect gay people from discrimination would seem increasingly urgent to the extent that minority sexual orientation is tacitly inferred from aspects of personal appearance that are routinely available for inspection.

In a related question, does my writing look gay?