We studied the success of predicting an epileptic seizure using two nonlinear indicators: the self-affinity index and the approximatе entropy of the EEG. We used the international University of Pennsylvania epileptic EEG database, which has records from electrodes implanted directly into the brain. For the analysis of two EEG indices in 5 patients, the records preceding an epileptic seizure were taken from electrodes implanted in the brain, under which an epileptiform EEG was observed. For comparison, we took records obtained from control electrodes. It turned out that in all five patients the approximation entropy of the EEG was significantly reduced compared to the control in those EEG records in which epileptic activity was subsequently observed. The self-affinity index revealed a decrease in only one patient. The results obtained indicate a more effective prediction of an epileptic seizure using the approximation entropy index.