This paper proposes an improved wavelet-based method for the delineation of T wave in the electrocardiogram by using multiscale differential operator (MDO). MDO was employed to automatically categorize the T wave into one of the three categories of T wave morphologies and to improve the precision of T wave delineation for each category. The new algorithm was evaluated on QT database (QTDB) and new annotations of 2160 beats from QTDB were performed by two cardiologists. To evaluate the performance of the wave delineation, the time differences between automatic detection versus cardiologists' annotations and intercardiologist differences were measured. The new algorithm can attain better results than the previous methods by achieving the smallest standard deviation of the differences and qualifying the strict error criterion for T-off measurement. This new algorithm also exhibited excellent T wave categorization agreement with the cardiologists, resulting in kappa values above 0.75.