We have previously reported that dogs can be trained to recognize specific changes preceding an epileptic seizure in humans. Such dogs can provide an overt signal that acts as a useful warning to the human. Early observations suggested that seizure frequency might also be reduced. We report a prospective study of 10 consecutive referrals to our Seizure Alert Dogs service of people with tonic-clonic seizures. Seizure frequency was monitored over a 48 week period including 12 weeks baseline after entry, a 12 week training period, and 24 weeks follow up. Comparing baseline seizure frequency to the last 12 weeks of follow up, there was a 43% mean reduction in seizure frequency ( P= 0.002). Nine out of /10 subjects showed a 34% or greater reduction, 4 /10 showed a 50% or greater reduction, and only one showed no improvement. Although a significant drop in seizure frequency was seen during the first 4 weeks of training ( P= 0.0078) a further drop occurred between the first and last 4 week period of training (P = 0.038) and this final improvement was maintained for the whole 24 week follow up.