The efficacy of antiepileptic drugs (AEDs) is partly determined by the type of seizure being treated. Some seizure syndromes may be improved by certain drugs, whereas others may be worsened. Therefore, drug trials, which recruit patients with different types of epilepsy syndrome may fail to show a global therapeutic effect. In order to demonstrate an effect, it is necessary to use a design of trial, which gives optimum power, and this may involve identifying as homogeneous a group of patients as possible. This 'splitting' approach permits a reduction in the number of patients included in a trial. It also allows identification of syndromes, which may be paradoxically worsened by a drug. It is necessary to review from time to time the diagnostic criteria for particular syndromes, otherwise they may prove to be too restrictive. At some point of drug development a 'lumping' approach is necessary in order to include patients whose clinical characteristics challenge any classification.