A ketogenic diet is used in children with drug-resistant epilepsy but predictors for efficacy are largely lacking. Our aim was to evaluate if causative genetic variants could predict seizure response to the ketogenic diet. A cohort study of 226 children with refractory epilepsy and classic ketogenic diet treatment for at least 3 months (76.9% of the 294 who started) was performed. The median age at diet start was 5.1 years (range 0.1-17.8), 118 were girls and 108 boys. They had previous trials of a median of 6.0 anti-seizure medications (range 0-12) and intellectual disability was found in 87%. Seizure response (≥50% reduction) was found in 138/226 patients (61.1%) at 3 months, 121 (53.5%) at 6 months, 107 (47.3%) at 1 year and in 80 (37.0%) at 2 years follow-up of ketogenic diet. Age of epilepsy onset was lower and combined epilepsy type less common in responders compared to non-responders but no differences were found for specific seizure types, ketogenic ratio or beta-hydroxybutyric acid blood levels. A causative pathogenic/likely pathogenic variant was detected in 107/153 = 69.9% in 48 different genes. Next generation sequencing was used in 91/226 (40%) cases with a diagnostic yield of 58.2% (53/91). In comparison with cases without a revealed genetic aetiology, patients with a causative genetic variant had less atonic seizures and epileptic spasms and a better seizure response with 17.3% seizure free and 25% with >90% seizure reduction at 2-year follow-up. Causative variants in , , and showed significant diet response ( < 0.05) and good efficacy was also associated with , , , , and . Causative variants in and were among genes linked to a lack of response. To our knowledge not described previously, we report a good ketogenic diet response related to causative variants in , , , , , , and but a lack of response for causative variants in , , , , , , , and After grouping of genes into functional groups, causative variants in transporter genes had the best response ( = 0.009) and variants in other membrane-related proteins (ion channels and neurotransmitter receptors) also showed good efficacy. However, the gene group related to cell structural integrity and/or homeostasis had the worst diet response ( = 0.00006). In conclusion, our results support that causative genetic variants may be used as prognostic markers of ketogenic diet response, constituting an example in the expanding area of precision medicine.