I didn't know the term "stereotype threat" and had to google it. Here is one of the clearer explanations I found:
"Black and White college students a half-hour test using difficult items from the verbal Graduate Record Exam (GRE). In the stereotype-threat condition, they told students the test diagnosed intellectual ability, thus potentially eliciting the stereotype that Blacks are less intelligent than Whites. In the no-stereotype-threat condition, the researchers told students that the test was a problem-solving lab task that said nothing about ability, presumably rendering stereotypes irrelevant. In the stereotype threat condition, Blacks did less well than Whites. In the no stereotype-threat condition, Blacks' performance rose to match that of equally skilled Whites."[1]
"Merely telling women that a math test does not show gender differences improved their test performance. The researchers gave a math test to men and women after telling half the women that the test had shown gender differences, and telling the rest that it found none. When test administrators told women that that tests showed no gender differences, the women performed equal to men. Those who were told the test showed gender differences did significantly worse than men, just like women who were told nothing about the test."[1]
[1] http://www.apa.org/research/action/stereotype.aspx
(I'm not saying that this effect is real. I'm just passing along an explanation of the term for the benefit of others.)