To investigate the epistatic interactions involved in antiepileptic drug (AED) resistance, 26 coding single-nucleotide polymorphisms (SNPs) were selected from 16 candidate genes. A total of 200 patients with drug-resistant localization-related epilepsy and 200 patients with drug-responsive localization-related epilepsy were genotyped individually for the SNPs. Rather than using the traditional parametric statistical method, a new statistical method, multifactor dimensionality reduction (MDR), was used to determine whether gene-gene interactions increase the risk of AED resistance. The MDR method indicated that a combination of four SNPs (rs12658835 and rs35166395 from GABRA1, rs2228622 from EAAT3 and rs2304725 from GAT3) was the best model for predicting susceptibility to AED resistance with a statistically significant testing accuracy of 0.625 (P < 0.001) and cross-validation consistency of 10/10. This best model had an odds ratio of 3.68 with a significant 95% confidence interval of 2.32-5.85 (P < 0.0001). Our results may provide meaningful information on the mechanism underlying AED resistance and, to the best of our knowledge, this is the first report of evidence for gene-gene interactions underlying AED resistance.