In order to extract the cardiac characteristics in electrocardiogram (ECG), a feature extraction technique was developed based on wavelet domain Lorentz differential deconvolution. During the feature extraction of QRS complex, baseline drifts were firstly removed from raw ECG records by a mathematical morphology method and the feature sub-band of QRS complex was separated by using wavelet transform. Then an evolving Lorentz differential deconvolution technique was applied to estimating the local features of QRS complex from this sub-band. During the feature extraction of P and T waves, the above steps were similarly employed and, before wavelet transform, QRS complex was eliminated through locating their positions to avoid relevant disturbance. The proposed technique achieved a recognition of 99.37% for QRS recognition and a detection rate of 98.62% for P waves detection when tested with the MIT/BIH Database. And validated with the QT Database, the results of QT intervals detection also showed an obvious improvement (85.26% when target domain less than 14 ms and 95.34% when target domain less than 28 ms separately on average).