Accurate and up-to-date land cover data are important for climate-change modeling. Quality assessment is becoming critical, as many satellite-based land cover products of differing scales have been released to meet the needs of scientific studies. In this study, we assessed the Moderate Resolution Imaging Spectroradiometer (MODIS) land cover products by analyzing the probability of interannual change from 2001 to 2012. The authors found that, cumulatively, 43.0% of MODIS land cover had changed over China from 2001 to 2012 at least once. Of this percentage, 12.1% was considered unreasonable change, 6.1% was considered reasonable change, and areas of confusion accounted for about 24.8%, giving rise to great uncertainty in the products. MODIS Collection 51 products clearly have less uncertainty than the Collection 5 products. Areas of reasonable change occurred in transition zones of ecological, biophysical, and climate gradients, while areas of unreasonable change appeared in heterogeneous landscapes. The misclassifications at three spatial scales of horizontal grids used in regional climate models occurred largely in the heterogeneous landscapes, and the areal percentage of misclassification decreased with larger horizontal grid spacing. In addition, the misclassifications in MODIS products often occurred among specific classes, which are geographically, ecologically, and spectrally similar, with low discriminative spectral-temporal signals. The effect of classification uncertainty should be made known, and further improvements are still needed for application in regional climate models. Our findings have important implications for better understanding the uncertainties of MODIS land cover products, and for improving the land surface parameterization for regional climate models.