While global warming during the last century has been well recognized, the magnitude of the climate warming in regions such as China over the past 100years still has some uncertainty due to limited observations during the early years. Several series of temperature anomalies for the 20(th) century in China have been independently developed by different groups. The uncertainty arises mainly from the sparse observations before 1950, where statistics are sensitive to the small and potentially biased sample. In this study, BSHADE-MSN (Biased Sentinel Hospitals Areal Disease Estimation and Means of Stratified Nonhomogeneous Surface), a combination of two novel distinct statistical methods that are applicable with different sample situations to a spatial heterogeneous surface, is applied to estimate annual mean temperature anomalies for China. This method takes into account prior knowledge of geographical spatial autocorrelation and nonhomogeneity of target domains, remedies the biased sample, and maximizes an objective function for the best linear unbiased estimation (BLUE) of the regional mean quantity. For the period 1900-1999, the overall trend estimated by BSHADE-MSN is 0.80 degrees C with a 95% confidential interval between 0.41 degrees C and 1.18 degrees C. This is significantly lower than that calculated by Climate Anomaly Method (CAM) and Block Kriging. The new temperature anomaly series for China exhibits slightly warmer conditions for the period before 1950 than existing studies. All the methods applied so far agree well with each other for the period after 1950, when there are sufficient stations across the country for the estimation of temperature anomaly series. Cross validation shows that the new regional mean temperature anomaly series has smaller estimation error variance and higher accuracy than those based on the other methods assessed in this study.