Lack of reproducibility in the scientific and lay literature of many scientific reports is an increasing concern, as are the high rates of failure to validate highly promising preclinical observations in clinical trials. There are many technical reasons why experimental results, particularly in cancer research, cannot be reproduced, including unrecognized variables in the complex experimental model, poor documentation of procedures, selective reporting of the most-positive findings, misinterpretation of technical noise as biological signal and, in the most extreme cases, fabrication of data. We suggest that cognitive biases in research and flaws in the academic incentive system also contribute to the publication of immature results. Recognition of these factors, which are often not discussed, provides additional strategies to improve reproducibility. We suggest that in addition to establishing better standards of data presentation and creating venues for publication of negative results, some changes to the grant submission and funding system could further improve the reproducibility of research findings.