We study whether racial or gender discrimination in marking exists at universities by conducting an experiment at a major Australian university where we randomly assigned names indicative of White, Chinese or Adopter identities (comprised of a White first name and Chinese surname) and male or female gender to real exam coversheets and recruited university graders to mark these exams. We find that the most economically-significant evidence of discrimination is found at grade thresholds. Exam scripts with Chinese and Adopter names are less likely than White names to receive a mark just above a grade threshold. Conversely, scripts with Chinese names receive a small marking bonus on average compared to the same script with a White name. Discrimination at grade thresholds is found to be more consistent with taste-based discrimination, whereas discrimination at the average is more consistent with statistical discrimination.