Background: Outcome reporting bias (ORB), whereby outcomes are selected for publication on the basis of the results, has previously been identified as a threat to the validity of meta-analysis. Objectives: To determine how accurately we could predict whether a review primary outcome was measured when it was not mentioned in a trial report. To determine how often ORB had occurred in trials. To investigate the impact of ORB in the cohort of reviews. Methods: The use of a nine-point classification system for detecting ORB in randomised controlled trials (RCTs) was developed and applied to a large unselected cohort of Cochrane systematic reviews. Trialists were contacted and the sensitivity and specificity of our method of ORB assessment was estimated. If the trialist confirmed that the review primary outcome data was measured and analysed, a reason for the non-reporting of the data was sought. The bound for maximum bias was used as a sensitivity analysis to assess the impact of ORB. Results: From an unselected cohort of 283 Cochrane reviews, the prevalence of reviews with suspected ORB was 55%. From a total of 48 trialists’ responses received to date, sensitivity results for detecting bias were high (88%), while specificity was reasonable (73%). The application of the maximum bias bound sensitivity analysis applied to 32 reviews revealed that the statistically significant conclusions of eight of these reviews were not robust to ORB. Conclusions: The presence of outcome reporting bias related to review primary outcomes is widespread. We show that bias can occur in meta-analyses as a result of ORB and show that the maximum bias bound sensitivity analysis is one of a number of potential solutions to the problem that will be discussed.