Vagus Nerve Stimulation (VNS) is recognized as an efficient procedure for controlling seizures in patients with drug-refractory epilepsies. It is used as a palliative procedure as a complement to conventional treatment by antiepileptic (AE) drugs and, according to literature, 40 to 50p.cent of patients report a decrease in seizures frequency >or=50p.cent, which is usually accepted to classify patients as responders in add on AE drug trials. The objectives of this study based on retrospective analysis of 50 consecutive patients with partial (39) or generalized (11) refractory epilepsy non eligible for surgery were; firstly to evaluate the global long term VNS efficacy and secondly to identify potential predictors of the VNS effects on seizure frequency. No patient has been seizure free at any moment of the follow up (2.8+/-1.8 years, max: 6 years) and the AE has been maintained in all. During follow up 44, 66, 61 and 58p.cent of patients were classified as responders at 6 months, 1, 2 and 3 years, respectively. Logistic regression analysis showed that: the percentage of responders at 6 months of follow up and later was significantly higher than that before 6 months (p=0.002); generalized epilepsy was predictive of a better outcome as compared to partial epilepsy (p=0.03); there was a trend for a better outcome in partial epilepsies symptomatic of a focal lesion than in those with normal brain MRI (p=0.06). These results are in line with previously published data in terms of global efficiency and confirm that seizures control does not reach its maximal level before at least one year of VNS. In severe generalized epilepsies (either secondary or cryptogenic) manifesting by frequent falls due to atonic or tonic-clonic generalized seizures VNS is a useful palliative procedure, which entails much les of surgical risk than callosotomy. The better VNS effects in patients with partial epilepsy possibly reflect the high incidence in our series of Malformations of Cortical Development, which have been identified as one the few variables possibly predictive of a response over 50p.cent of seizures frequency reduction.