There has been a series of high-profile failures of drugs in clinical trials of acute ischaemic stroke that were designed to meet criteria necessary for drug regulatory approval. This has, again, called into question the value of animal models for identifying effective neuroprotective agents. Here, we review evidence that physiological changes (reperfusion, hyperglycaemia, hypothermia and blood pressure) produce comparable changes in outcome in both animal models and human stroke patients, which indicates that the models should identify clinically effective neuroprotective agents. We suggest that most clinical failures have occurred because compounds were administered differently in animal and clinical studies. We review earlier guidelines on the information that is necessary from preclinical studies before a compound enters clinical trials, and propose modifications to these guidelines.