We conducted a retrospective survey of the efficacy of ketogenic diet therapy (KD) for childhood refractory epilepsy. The subjects were 54 patients (male: 19; female: 35) undergoing KD between 1984 and 2007 in our hospital. We defined marked improvement as being seizure-free and satisfactory improvement as a reduction of the seizure frequency by more than 50%. KD was started at ages ranging from 6 months to 15 years (median:4 years), and maintained for 1 month to as long as 7 years. Twenty-nine patients (54%) discontinued KD within 1 year of its initiation. As the results, 18.5 and 35% of the patients showed marked and satisfactory improvement, respectively, at 1 year after starting KD. A greater than satisfactory improvement was achieved in 80% of those showing transitional state from West to Lennox-Gastaut syndrome (n = 5), 100% of those with epilepsy with both focal and generalized seizures (n = 3), 46% of those with severe myoclonic epilepsy in infants (n = 13), 33% of those with cryptogenic myoclonic-astatic epilepsy (n = 3), 25% of those with symptomatic localization-related epilepsy (n = 8), 20% of those with Lennox-Gastaut syndrome (n = 5), and 12% of those with symptomatic generalized epilepsy (n = 17). There were no serious side effects. Five patients discontinued KD despite their satisfactory improvement because they disliked the taste of the meals. KD therapy is worth considering for children with refractory epilepsy, but a more palatable KD menu needs to be developed for Japanese patients to maintain their long-term compliance with the therapy.