Predicting the ecological causes and consequences of global climate change requires a variety of approaches, including the use of experiments, models, and surveys. Among experiments, mesocosms have become increasingly popular because they provide an important bridge between smaller, more tightly controlled, microcosm experiments (which can suffer from limited realism) and the greater biological complexity of natural systems (in which mechanistic relationships often cannot be identified). A new evaluation of the contribution of the mesocosm approach, its potential for future research, as well as its limitations, is timely. As part of this review, we constructed a new database of over 250 post-1990 studies that have explored different components of climate change across a range of organisational levels, scales, and habitats. Issues related to realism, reproducibility and control are assessed in marine, freshwater, and terrestrial systems. Some general patterns emerged, particularly at the ecosystem level, such as consistent and predictable effects on whole-system respiration rates. There are, however, also many seemingly idiosyncratic, contingent responses, especially at the community level, both within and among habitat types. These similarities and differences in both the drivers and responses highlight the need for caution before making generalisations. Finally, we assess future directions and prospects for new methodological advances and the need for greater international coordination and interdisciplinarity. © 2013 Elsevier Ltd.