Abstract: In order to improve the accuracy of electromagnetic ultrasonic testing signal, a data processing method based on improved EMD was proposed. Aiming at the problem of losing high-frequency useful information in IMFs screening, an Euclidean distance based IMFs component screening criterion was established, and wavelet threshold filtering method was combined to achieve noise removal and effective extraction of small useful signals. In order to verify the effectiveness of the algorithm, thickness measurement experiments were carried out on 10 mm and 15 mm aluminum plates. The detected data were processed by using the improved EMD algorithm and the peak position was extracted. The thickness value was calculated by the time difference between adjacent peak points. Multiple groups of experiments were carried out and the error was calculated. The experimental results show that the proposed method can improve the accuracy of electromagnetic ultrasonic testing, and improve the signal-to-noise ratio of echo signal. The minimum error can reach 0.1% for 10 mm aluminum plate and 0.3% for 15 mm aluminum plate.