In recent years, most of the published trials of new antiepileptic drugs (AEDs) have had a parallel-group design rather than cross-over. Nevertheless, in some situations the cross-over trial does have advantages, and therefore, its role needs re-appraisal. The crossover design requires a much smaller number of patients for a similar statistical power because patients act as their own controls, which is a particular advantage when the type or severity of epilepsy varies widely in the patients recruited. As a result, the financial cost is smaller and fewer patients are exposed to the new agent, perhaps with ethical arguments in favour of this type of design in proof-of-efficacy trials. Within-patient analysis also makes the detection of drug interactions more robust. On the other hand, there is a theoretical risk that the beneficial effects of the first treatment (or conversely, withdrawal seizures on stopping it) might carry over into the second treatment period and thereby confound the detection of treatment effects. The parallel-group design is more versatile in that a stable disease state is not a pre-requisite, and therefore, trials in newly diagnosed patients are possible. Multiple treatment limbs are also more practical. The duration of a parallel-group trial may be shorter because only one treatment period is involved, although this may be offset by the much larger number of patients needed to be recruited and the time involved in doing so. Parallel-group trials almost always require a multicentre approach, with the inevitable logistic problems involved. It is argued that proof-of-principle and add-on proof-of-efficacy trials of a new drug are more efficiently undertaken using a cross-over design but that subsequent evaluation will require the versatility of trials with a parallel-group design.