Obtaining yield distribution map of farmland and analyzing the spatial difference of plot yield are important foundations for implementing precision farming. In order to accurately collect the spatial difference of yield, and at the same time, the acquisition accuracy of yield monitor system and the interpolation accuracy of the output spatial distribution map are improved. The selfdeveloped realtime monitoring system of harvester was used. Based on accurate yield spatial distribution map, spatial variability analysis was conducted on wheat yield data from 2013 to 2015. Firstly, the results showed that the pretreatment method of threshold filtering can effectively eliminate outliers and restore the real yield distribution. Secondly, by comparing the RMSE values, it was determined that spatial distribution map of experimental plot yields drawn by the ordinary Kriging (OK) had higher interpolation accuracy. The minimum value of 826.70kg/hm2 appeared in the index mode of OK method for 2013, search strategy was elliptical, the largest adjacent element was 5, the smallest adjacent element was 3 and 1 sector. Finally, the curve parameters of semivariance function were used to obtain the spatial variability information of three seasons and optimal sampling interval of the system. The spatial variation of yield in 2013 and 2014 were entirely caused by spatial autocorrelation. And that of 2013 was mainly in the mesoscale range of 2~12m, and that of 2014 was in the mesoscale range of 2~5m. The spatial variation caused by random factors in the 2015 was 25%, which was in the small scale range below 2m. The spatial autocorrelation caused variation of 75%, which was in the mesoscale range of 2~15m. The sampling interval of the system should be kept at 2~10m. Too small or too large pitch was affected by large random factors or reduced interpolation accuracy. These results can be used to develop fine management decisions for farmland.