When we rely on science to inform decisions about matters such as the environment, teaching strategies, economics, government, and medicine, evidence-based decision-making can only be as reliable as the totality of the science itself. We must avoid distortions of the scientific literature such as publication bias, which is an expected systematic difference between estimates of associations, causal effects, or other quantities of interest compared to the actual values of those quantities, caused by differences between research that is published and the totality of research conducted. Publication bias occurs when the probability of publishing a result of a study is influenced by the result obtained. It appears to be common and can produce misleading conclusions about interventions, make effects appear greater than they are, lead to irreproducible research, and ultimately undermine the credibility of science in general. Methods to detect publication bias and steps to reduce it are discussed.