Signal analysis methods for EEG signal are applied in blind signal separation, artifact removal, and feature extraction. These methods are utilized in time-domain or frequency-domain. A fundamental problem in EEG signal is that signal from the scalp exhibits nonstationarity and is mixed spatiotemporally. In this study, we analyzed EEG signal using a combination method with ICA(Independent Component Analysis)and semiparametric approach. Theta is associated with creative inspiration, deep mediation, and attention. During the episodic memory retrieval, the combination method is used to analyze data of7channels, in which EEG data from6electrodes and one EMG data are recorded every5ms. Subjects are required to push a right-side button when they recognized a presented word-pair is the same one as they previously remembered. Three seconds length EEG data are recorded before and after button pressed, then the artifact (eye blinking)is removed by ICA. Our proposed combination method is more effective than parametric AR model in context of extracting theta activity(4Hz to8Hz)during the episodic memory retrieval.