Very few trials of acute stroke treatments show efficacy of a tested agent on the prespecified primary outcome. We can learn many lessons from the studies that achieve only neutral results. Preclinical studies have been flawed by use of models of transient not permanent brain ischaemia, treatments that aim to protect cerebral grey matter independently of white matter, delivery of the study drug within too short a time window after ischaemic insult, use of surrogate outcome measures in the short term instead of function in the long-term, and small sample sizes. Clinical trials have been hampered by heterogeneity in causes of stroke and inability to classify subtypes of cause; the short time available to rescue ischaemic brain tissue; the haemorrhagic transformation that can cause severe functional consequences seen frequently in infarcted brain tissue; the lack of valid, reliable, sensitive, and simple tools for assessment of functional outcome; and, above all, small treatment effects that are difficult to detect or refute. In this review, we look at the designs and results of all controlled trials of treatments for ischaemic stroke, and try to identify opportunities to improve future treatment assessment.