The study on many years’ land cover plays a crucial role in promoting the high-quality development of the Yellow River basin. Meanwhile, high-frequency and high-precision land cover data are vital for land cover monitoring. This study took the basin’s geometric center that has been stable for many years to sample and quickly selected a set of sample points that can be used for annual image supervised classification. Then, cloudless images were screened out from nearly one thousand Landsat images on average of the Yellow River basin of each year from 2000 to 2020 and were spliced by year using Google Earth Engine. Then, the random forest classification method was used to conduct the supervised classification of the cloudless images, producing the annual land cover data of the Yellow River basin in the recent 20 years. Finally, the land cover data of 2010 of the basin were compared with well-known annual land cover data at home and abroad. The results are as follows. ① The selection method of sample points used in this study is reasonable and reliable, with a selection accuracy of more than 94.7%, meeting the requirements of sample accuracy for supervised classification. ② The overall accuracy of the annual land cover data created based on Google Earth Engine is 0.82±0.03, with an average Kappa coefficient of 0.82. The classification accuracy and the overall and local classification results are better than the MCD12Q1 and ESA-CCI datasets. ③ Using the method for creating annual land cover data using Google Earth Engine, the frequency and accuracy of large-scale land cover data can be considered at the same time to a certain extent.