Arrhythmias can cause abnormalities in the radial pulse wave of the patient’s wrist. In this paper, the differences between arrhythmia pulse wave and healthy pulse wave in morphology, spectrum and wavelet energy are analyzed. We use the mean period (MT) of the pulse wave, the standard deviation of the period (MST), the harmonic frequency (f i ), the harmonic amplitude (fp i ), the spectral energy ratio (sen), the -wavelet energy (E i ), and the -wavelet entropy (WE) as the features of the pulse wave, and the support vector machine was used to classify the arrhythmia pulse wave and the healthy pulse wave. The results show that when three feature combinations of morphological features (MF), spectral features (SF), and -wavelet energy features (WF) are selected, the performance of the classifier is the best, and the accuracy rate is 95.18%, and the corresponding sensitivity is 96.08%, specificity was 93.75%, F1_scor was 0.9608, and AUC value was 0.9871. The research results show that the three characteristics of MF, SF, and WF can be used as objective indicators to distinguish arrhythmia pulse wave from healthy pulse wave. Combined -with machine learning method, it provides a new auxiliary evaluation index for clinical diagnosis of arrhythmia. The method of arrhythmia identification by pulse wave is feasible.