The electrocardiogram (ECG) reflects the cardiac electrophysiological activity, which is essential for the automatic diagnosis of heart disease. In this article, an automatic classification method of arrhythmia using multi-scale recurrence plot and vision transformer is proposed. Firstly, the ECG is decomposed into low-frequency components and several high-frequency components using wavelet transform, which are respectively transformed into 2D texture image by the recurrence plot method. To solve the problem of sample imbalance, the conventional vision transformer is improved by replacing cross entropy loss with multi-classification focal loss. Finally, the arrhythmia classification is performed by utilizing the multi-scale recurrence plot representation of ECG and the improved vision transformer. The MIT-BIH arrhythmia dataset is utilized to evaluate the proposed arrhythmia classification method. The average accuracy of the proposed method is 97. 38% . Experiment results show that the proposed method is effective and better than other conventional method.