The electroencephalogram (EEG) is an important tool in determining the presence or absence of clinical seizures in neonates. The design of an efficient automated EEG seizure detection method suitable for use in the clinical environment would be beneficial. Existing methods include a time-frequency (TF) matched filter approach, which correlates TF seizure templates with time-frequency distributions (TFDs) of the EEG to detect seizure. A major problem with this method is the pre-selection of a template set that adequately represents all different TFD seizure types. To overcome this problem, a new method is proposed that applies the TF matched filter but eliminates the need for a predefined template set. Preliminary results are presented for both real and simulated EEG data.