Abstract: Abstract: In order to obtain real time grain yield data information, a kind of grain yield monitor system based on photoelectric principle was developed. It was consisted of sensor module, data collection module, GPS module and grain yield calculation terminal. The optical reflectance type grain volume sensor was installed on one side of the combine elevator. When the grain was conveyed through the grain flow sensor, the scraper and grain would be block the light path, intermittently. As a result, a pulse width signal would be generated. And the pulse width signal was proportional to the thickness of scraper and grain volume. Other sensors signals such as elevator speed sensor signal and GPS signal would also be generated at the same time. After the data collection module, all the signals would be transmitted to the liquid crystal display (LCD) terminal by RS485 bus. The grain volume monitor software could display grain instantaneous output volume, grain production information, harvest area and other information. After analyzing the working status of combine harvester and the simulation of scraper heap shape, a subsection type grain yield transformation model was proposed. As the accuracy of grain yield monitor system was affected by the elevator speed seriously, the model had also considered the elevator speed as an input parameter. When the combine harvester worked at the normal status, grain volume had linear relationship with scraper grain thickness. In order to further optimize the quality of yield data, a new preprocessing method was also proposed based on elevator speed dynamic threshold value filter. The experiment selected a field of 2.67 hm2 Xiaotangshan National Demonstration Station in Beijing, located in E116?26?54.66?-116?27?02.41?, N40°11′15.50″-40°11′11.61″. The grain yield monitor system was installed TB60 (4LZ-6 B) type self-propelled combine harvester produced by Zoomlion Corporation. According to the combine elevator in different speed conditions of no-load scraper thickness, the upper and lower bounds of normal production were determined as the original data of singular value standard. Once the data was below 10% of the real time calculated scraper thickness, it was removed. Once it was above 5 times of the real time calculated scraper thickness, it was replaced by the normal value one second before. In order to evaluate this new preprocessing method, original data, average filter data and dual threshold filter data were used to validate the model. The test results showed that the proposed data preprocessing method could eliminate the singularity and improve the smooth of yield data, obviously. The coefficient of variation (CV) also decreased to 0.33 from 0.53. The field experiment showed that validation error of the grain yield monitor model was less than 3.50%, which could satisfy the practical need. Compared with foreign similar type grain yield monitor system, such as SMARTYIELD Pro system produced by Raven Corporation and the Yield Monitor System produced by Trimble Corporation, the developed system had the advantage of simple calibration step and convenient installation method.