There is room for improvement in how science is done and reported, but something can often be learned from irreproducible experiments. The situation may not be as dire as some headlines imply.

It is crucial to include caveats when citing analyses of reproducibility. For example, an often-quoted 2015 survey of factors that could improve the reproducibility of scientific results (see go.nature.com/yxwgmb) noted that there was a low response rate to the questionnaire, a qualifier that is not always mentioned.

It is important to recognize that researchers cannot control for an unknown variable. Take a web tool for identifying unwanted 'passenger mutations' that could confound analyses of transgenic mice (T. Vanden Berghe et al. Immunity 43, 200–209; 2015). This tool arose from reports of mouse phenotypes that, unbeknown to researchers, depended on unintended mutations. This is an example of a useful resource that enhances our understanding of underlying biological phenomena and results from experiments that might otherwise be branded as irreproducible.

It is in this context that scientific societies are pushing to increase experimental rigour and reporting transparency. For instance, guidelines from the Federation of American Societies for Experimental Biology (go.nature.com/zdf89b) aim to help scientists to meet the reproducibility requirements of research funded by the US National Institutes of Health.