Abstract: In order to obtain the fatigue load spectrum of rotary cylinder,the service load of rotary cylinder was ob- tained using large capacity data recorder (black box). Amplitude threshold method was used to remove abnormal peak points in the data. Pretreatment was carried out on the measured load to obtain the sample data of the rotary cylinder pull load after the test of load stability and ergodicity. The rotary cylinder service load sample data was statistically pro- cessed using the rainflow counting method. Combined with a lot of statistical analysis experience,the Weibull distribu- tion of amplitude value and normal distribution of mean value were assumed respectively. The distribution law of its amplitude value and mean value was studied,and the independence of amplitude value and mean value was tested by Chi-square test. And the three-modal normal probability distribution model was proposed for the mean value. The re- sults show that the amplitude and the mean of rotary cylinder pull load are independent in the condition of 0. 05 test level. The amplitude accords with the Weibull distribution with a shape parameter of 1. 156 and a scale parameter of 3. 345. The proposed three-modal normal probability distribution model reflects the statistical characteristics of the mean and realizes the accurate identification of the load information of rotary oil cylinder. In order to keep the working load characteristics of the rotary cylinder in the whole life cycle,the cumulative frequency of rotary cylinder pull load sample data was extended to 5×105 cycles according to the proportion of rotary cylinder operating time,the two-dimen- sional load spectrum with 8×8 levels of rotary cylinder in pull condition was established using two-dimensional proba- bility distribution function. Based on fatigue cumulative damage theory,two-dimensional load spectrum changed into one-dimensional load spectrum with 8 levels of the rotary cylinder pull load using the Goodman method,which can pro- vide the basis of the fatigue life prediction and fatigue life testing.