Clinicians who manage delirium must do so without key information required for evidence-based practice, not least lack of any clearly effective treatment for established delirium. Both the nature of delirium and the methods used to research it contribute to difficulties. Delirium is heterogeneous, with respect to motor subtype, aetiology, setting and the co-existence of dementia, and may be almost inevitable towards the end of life. Elements of assessment are subjective, so diagnosis can be uncertain or unreliable. Defining objectives of care and outcomes is sometimes unclear. Better identification and case definition, including seeking biomarkers, stratification by type, or aetiology, and application of more complex models of causation may help. This will likely require further observational epidemiology, imaging and laboratory-based research before further rounds of large-scale randomised controlled trials. Application of trial methodologies designed for drug treatments of better-defined conditions may have failed to take account of the complexities both of diagnosis and complex intervention in delirium. Both drug and complex intervention trials need sufficient preliminary work to ensure that the right dose, duration or intensity of treatment is delivered and a range of 'intermediate' and 'distal' outcome measures assessed. Re-purposing of established drugs may provide a source of investigational products. Greater use of alternative research methodologies (qualitative and realist), or adjuvants to trials (process evaluation), will help answer questions about focus, generalisability and why interventions succeed or fail. Delirium research will have to embrace both a 'back to basics' approach with increased breadth of methodologies to make progress.