The prognosis for children diagnosed with epilepsy is generally good, with a large proportion responding well to the first treatment given. A substantial proportion, however, will not respond well to treatment, and for these patients the clinical goal is to find an optimal balance between the benefits and side-effects of any treatment given. For the newly, or recently, diagnosed population, the key question for the newer drugs is how soon they should be tried. The cost-effectiveness of using these agents early, in place of one of the older agents, will depend on the effectiveness and tolerability of these agents compared with the older agents; the evidence from the available trial data so far suggests that the newer agents are no more effective but may be somewhat better tolerated than the older agents, and so the cost-effectiveness for early use will depend on the trade-off between effectiveness and tolerability, both in terms of overall (long-term) treatment retention and overall utility associated with effects on seizure rate and side-effects. There are insufficient data available to estimate accurately the nature of this trade off either in terms of long-term treatment retention or utility. Better information is required from RCTs before any rational evidence-based prescribing strategy could be developed. Ideally, RCTs should be conducted from a 'public health' perspective, making relevant comparisons and incorporating outcomes of interest to clinicians and patients, with sufficiently long-term follow-up to determine reliably the clinical utility of different treatments, particularly with respect to treatment retention and the balance between effectiveness and tolerability. RCTs should mirror clinical practice with respect to diagnosis, focusing on defined syndromes or, where no syndrome is identified, on groups defined by specific seizure type(s) and aetiology. Epilepsy in children is a complex disease, with a variety of distinct syndromes and many alternative treatment options and outcomes. Diagnosis-specific decision-analytic models are required; further research may be required to inform parameter values adequately with respect to epidemiology and clinical practice.