Abstract Background Valproate is a first-line treatment for newly diagnosed idiopathic generalised or difficult to classify epilepsy, but not for women of child-bearing potential due to teratogenicity. Levetiracetam is increasingly prescribed despite lack of evidence of clinical or cost effectiveness. Methods This randomised un-blinded controlled trial compared starting treatment with levetiracetam or valproate for generalised or unclassified epilepsy. Adult and paediatric neurology services across the UK recruited participants aged 5 years or older with two or more unprovoked generalised or unclassifiable seizures. SANAD II was designed to assess the non-inferiority of levetiracetam compared to valproate for the primary outcome time to 12 month remission. The non-inferiority limit was a hazard ratio (HR) of 1·314, which equates to an absolute difference of 10%. HR > 1 indicates an event is more likely on valproate. Findings 520 participants were recruited between April 2013 and August 2016 and followed up for a further 2 years; median age 13.9 (range 5.0 – 94.4), 64.8% male, 397 participants had generalised and 123 unclassified epilepsy. Levetiracetam did not meet the criteria for non-inferiority in the intention to treat analysis of time to 12-month remission HR 1·19 (95% CI 0·96-1·47); non-inferiority margin 1·314. The per-protocol analysis of time to 12-month remission found valproate superior to levetiracetam HR 1·68 (95% CI 1·30 - 2·15). Valproate was also superior to levetiracetam for times to 24-month remission HR 1·43 (95% CI 1·06 - 1·92), first seizure HR 0·82 (95% CI 0·67 - 1·00), treatment failure HR 0·65 (95% CI 0·50-0·83) and treatment failure due to inadequate seizure control HR 0·43 (95% CI 0·30 - 0·63); treatment failure rates due to unacceptable adverse reactions were similar HR 0·93 (95% CI 0·61 - 1·40). Adverse reactions were reported by 37·4% participants randomised to valproate and 41·5% participants randomised to levetiracetam. Levetiracetam was dominated by valproate in the cost-utility analysis, with a negative incremental net health benefit of -0·040 (95% Central Range (CR) -0·175, 0·037) and a probability of 0·17 of being cost-effectiveness at a threshold of £20,000 per QALY. Interpretation Compared to valproate, levetiracetam was found to be neither clinically nor cost effective. For girls and women of child-bearing potential these results inform discussions about benefit and harm of avoiding valproate.