There is great concern that results published in a large fraction of biomedical papers may not be reproducible. This article reviews the evidence for this and considers some of the factors that are responsible and how the problem may be solved. One issue is scientific fraud. This, in turn, may result from pressures put on scientists to succeed including the need to publish in “high impact” journals. I emphasise the importance of judging the quality of the science itself as opposed to using surrogate metrics. The other factors discussed include problems of experimental design and statistical analysis of the work. It is important that these issues are addressed by the scientific community before others impose draconian regulations.