A prominent 2011 paper in Science found that white researchers receive grants from the US National Institutes of Health (NIH) at nearly twice the rate that African American researchers do (D. K. Ginther et al. Science 333, 1015–1019; 2011). Although some of the disparity could be explained by differences in education, institution and publication record, the sheer magnitude of the result seemed to suggest that something more insidious was at play.

The idea that scientists who volunteer time and energy to review NIH grants could be biased against qualified minority researchers is a tough pill to swallow. The NIH is to be commended for not sweeping this possibility under the rug: it has turned to the scientific method to investigate the suggestion. Over the next three years, researchers will strip names and other identifying information from grant applications, text-mine proposals for subtle clues to an applicant’s race that might subconsciously tip off reviewers, and study reviewer critiques to see whether they hold evidence of prejudice (see page 243).

It is a topic that the NIH will need to broach delicately. Few academics consciously hold any such inclinations, and fewer still would deliberately allow them to affect their grant evaluations. Some are likely to bristle at what might be seen as an accusation of racism, and the NIH plans to conduct at least some of its studies of grant reviews without the reviewers’ knowledge or consent.

But better for the NIH to offend a few people than to make snap judgements and institute blunt policies to address the problem. Fixes such as increasing scholarships and training for minority groups would no doubt be a good thing, but they could be an unhelpful use of money if they do not address the root cause of the disparity. And policies such as grant-allocation quotas could come at the expense of other researchers.

The NIH says that it will be guided by the data produced. Interventions could include training programmes on bias for reviewers, or using peer reviewers from different demographic groups, such as early-career scientists. If the agency does find evidence of bias, fixing it will be a difficult task. Unconscious bias, wherever it resides, is a difficult thing to turn off, even for the most educated and progressive of people.