Aiming at the problems of high cost, complex structure and low stability of grain combined harvester yield monitoring system, a grain harvester yield monitoring system was designed based on duty cycle measurement, which consisted of photoelectric sensor, GPS module, data processing unit, data storage unit and visualization unit. When the system worked, two voltage signals of grain covering and non-covering on the scraper were monitored by the reflection photoelectric sensor, and the highly corresponding duty ratio in the signal was processed by the software system. The relationship between the duty ratio and the output measurement model was used to obtain the output data, which was stored in the system together with the time and GPS data of the system. Through EDEM simulation and theoretical model analysis, the direct proportional relationship between the measured duty ratio and grain mass was deduced. The global model and local model were fitted to the measured duty cycle and grain mass by bench test, and the R2 of all the fitting lines were not less than 0.988. Then the global model and the local model were analyzed by the bench test. The bench test results showed that although the local model may be better for the measured data at fixed speed, the global model was more versatile. With the increase of measurement data, the relative error was decreased gradually. In the field experiment, the abnormal signals in the system measurement process were counted and analyzed. In order to reduce the influence of abnormal signals on the measurement error, the measured values of the system were calibrated with the actual output. The field experiment results showed that the maximum relative error of the yield monitoring system was 3.83%, the average relative error was 0.40%, and the overall error and error fluctuation of the system were small.