Abstract: Aimed at solving the problem that the traditional engineering methods of determining the load sample size is difficult to adapt to the field test, a method of determining the load sample size based on the Dynamic Time Warping (DTW) distance was proposed to calculate the agricultural machinery field test load sample size. The DTW distance method is mainly applied in voice recognition and online signature verification. It has the characteristics of being able to judge the similarity of unequal length time series and not relying on data statistic features for data similarity judgment. These features are more suitable for calculating the load sample size of agricultural machinery. Firstly, a dynamic load test system was built to obtain the torque load of the tractor drive shaft in the process of ploughing, and the small load data which was lower than 10% of the maximum load cycle was filtered. According to characteristics of the agricultural machinery operation and the tractor drive shaft load, the ploughing condition and the adjustment condition were regarded as a complete operation cycle, which ensured that these operation cycles meet ergodicity test. Moreover, traditional approximate mean accuracy estimation method and fitting curve of mean method were selected to determine the load sample size of the tractor drive shaft under ploughing condition and the results were 8 subsamples and 3 subsamples, respectively. After normalizing the load time series with different numbers of operation cycles, the DTW distance between them and the maximum number of operating sections were calculated. The results displayed that load sample size determined by DTW distance method was 16 subsamples. Furthermore, since the actual measured load cannot represent the whole life load of components, in practice, the measured load samples need to be statistically analyzed and then combined with the extrapolation formula to extrapolate the load to obtain the full life load spectrum. The accuracy of the whole life load spectrum is mainly related to the extrapolation method and the selected sample length. Therefore, under the same error condition, the commonly used parameter extrapolation was selected to verify the load sample size. On the premise that the confidence level was 95% and the goodness of fit is greater than 0.9, the fitting parameters of the double normal distribution function about the mean load and its frequency were obtained. The value of fitting function obtained from the whole measured load used as the reference value, and the relative errors between the reference value and the above fitting parameters were calculated. The results showed that the relative errors of the fitting parameters through the load sample size with the DTW distance method could be guaranteed to be within 10%. Using the double normal distribution function for mean load and its frequency fitting, the maximum relative errors of fitting parameter for the approximate mean precision estimation method and the mean curve were 126.06% and 80.62%, respectively, while the maximum value of the relative error of the fitting parameters based on the DTW distance calculation method was 5.43%, which verified the applicability of the load sample size calculation method based on the DTW distance. The calculation method can be used for the field test load size of tractor drive shafts in the future.