Automated seizure detection is important for speeding up epilepsy diagnosis or for controlling an implantable brain stimulator to avert seizures. Various features calculated from the electroencephalogram (EEG) can be used to detect seizures, and combining features can give superior detection performance. This paper investigates the correlation between seizure detection features in order to determine which ones should be combined for the purposes of seizure detection. Combinations of three features involving relative average amplitude, relative scale energy, coefficient of variation of amplitude, relative power, relative gradient and bounded variation tended to show the lowest correlations.