Background: Stereotype threat research revealed that negative stereotypes can disrupt the performance of persons targeted by such stereotypes. This paper contributes to stereotype threat research by providing evidence that domain identification and the difficulty level of test items moderate stereotype threat effects on female students' maths performance. Aims: The study was designed to test theoretical ideas derived from stereotype threat theory and assumptions outlined in the Yerkes-Dodson law proposing a nonlinear relationship between arousal, task difficulty and performance. Sample: Participants were 108 high school students attending secondary schools. Method: Participants worked on a test comprising maths problems of different difficulty levels. Half of the participants learned that the test had been shown to produce gender differences (stereotype threat). The other half learned that the test had been shown not to produce gender differences (no threat). The degree to which participants identify with the domain of maths was included as a quasi-experimental factor. Results: Maths-identified female students showed performance decrements under conditions of stereotype threat. Moreover, the stereotype threat manipulation had different effects on low and high domain identifiers' performance depending on test item difficulty. On difficult items, low identifiers showed higher performance under threat (vs. no threat) whereas the reverse was true in high identifiers. This interaction effect did not emerge on easy items. Conclusions: Domain identification and test item difficulty are two important factors that need to be considered in the attempt to understand the impact of stereotype threat on performance.