The automatic spike detection in EEG is significant in both diagnosing epilepsy and alleviating the heavy labor force of the doctors. This paper proposes an empirical model decomposition (EMD) based epileptic spike detection method. It extracts the high frequency components related to spikes in EEG signal by EMD, and it detects the spikes by calculating the instantaneous amplitude of the high component with Hilbert transform. The results of experiments show that the method works well.