The call by Glenn Begley and Lee Ellis to raise standards in preclinical cancer research (Nature 483, 531–533; 2012) is an admirable piece of self-critique that can be extended to all who study and treat human disease. Unfortunately, the kernel of their polemic — that reproducing published results can be problematic — was widely misinterpreted by the media.

The authors highlight an investigation in which scientific findings were confirmed in only 6 of 53 cases. This was picked up by news agencies, including Reuters, many of which construed the meaning as 'findings could not be replicated' — without mentioning the authors' nuanced caveats regarding the endpoints tested.

The public cannot be expected to appreciate the implications of, for example, a western blot working in one lab and not in another. Therefore, the meaning of 'reproducibility' in a particular context should be properly articulated. This could have prevented the mistaken inference that '90% of all science cannot be reproduced'.

In the complex data sets now being generated by genomics and proteomics, the nature of reproducibility is changing; it is increasingly dissimilar from that for more simple measurements. The basic meaning of scientific knowledge is shifting, and the community must learn to deal with it in ways that go beyond simple semantics.