Global land cover maps (GLC) are essential input data for many scientific studies, so assessment of their category accuracy and category confusion is very important for some specific applications. In this paper, the authors chose China as the study region and FROM, MODIS, GLOBCOVER and ESACCI as land cover data for validation. The authors first aggregated the four GLC and referenced data provided by some international organizations into eight categories, and then validated four products through the category consistency and confusion matrix in national scale. The relative comparison between FROM, MODIS, ESACCI and GLOBCOVER shows that the four land cover products have the similar category constituent. Forest, grassland, cropland and bare land are the major land cover categories, whereas shrub, build up and water/wetland are relatively rare. Through comparing one by one between referenced data and land cover products, the authors constructed the confusion matrix, and the validated results demonstrate that FROM and MODIS have the best overall agreement with referenced data at national scale; for example, FROM’s overall accuracy is 0.69, and MODIS is 0.67, and ESACCI’s overall value is 0.65. Conversely, GLOBCOVER has the worst overall accuracy, with the value being only 0.55. Forest, cropland, built up land and bare land all have the better category accuracy, so each of them would be as input data for national forest inventory, food security and urban expansion, but shrub's category accuracy is low in four global land cover products, with confusion mainly occurring with forest, grass and cropland . The study results not only provide some scientific reference for selecting the input data for ecological environment modeling, land cover change analysis, natural resource survey, but also provide a reasonable advice for the research direction in future land cover mapping projects.