Chinese cancer registration data provide information on the national cancer burden but how representative they are of the real situation is uncertain. Mortality data from cancer registration and the third national death survey were therefore compared to determine the accuracy of estimates in China. The data were from the Cancer Registration Annual Report, 2004, China and the third National Death Survey, 2004-205. Negative binomial regression was used to estimate site-specific cancer mortality rate ratios between the two. The estimated cancer mortality for all sites from national cancer registration was representative for China, especially in urban areas, but mortality was over-estimated for rural areas, with large differences in some cancer site-specific mortalities. Although cancer registration data are representative at the country level and for urban areas, they may not reflect real cancer burden in rural areas, particularly with some cancer types. Setting up new cancer registries in non-high risk areas in rural areas should be enhanced in further cancer surveillance plans.