The spatial production allocation model (SPAM) is one of the broadest spatial models of crop distribution and applies a cross-entropy method to downscale the global area and yield for multiple crops in the years 2000 and 2005 with a resolution of 5 arc min. To evaluate the allocation accuracy of SPAM for three staple crops (rice, wheat, and maize) in China, we compared these crop maps with remote-sensed cropland data derived from national land cover datasets. This comparison was conducted using a scheme that accounts for spatial differences at the pixel level. Overall, the map of maize has the highest area accuracy, with 64% reasonable pixels (covering 96% of the total maize area); these values were 57% (90% coverage) and 44% (81% coverage) for the wheat and rice maps, respectively. On the provincial scale, the area accuracies of crop maps in the top 10 provinces are better than those of the other provinces. Furthermore, the crop area consistency in rain-fed cropland is better than that in irrigated cropland. These evaluations provide decision makers with information regarding the strengths and weaknesses of SPAM products. This study also recommends priorities for further work to improve the reliability, utility, and periodic repeatability of crop distribution products.