This paper focuses on the estimation of land surface temperature (LST) based on the measurements from the visible and infrared radiometer (VIRR) and the medium resolution spectral imager (MERSI) on board the second-generation polar-orbiting meteorological satellite of China (FY3). The moderate spectral resolution atmospheric transmittance algorithm and computer model (MODTRAN) were used to simulate FY3 VIRR and MERSI infrared data. The split-window algorithm was employed to build LST estimation models based on the simulated data. The LST estimation experiments were conducted in Jiangsu Province. Nine images with fewer clouds cover were acquired on January 23 and February 3 and 11, 2012. They consisted of three VIRR band 4 images, three VIRR band 5 images and three MERSI band 5 images, which were used to retrieve LST of the experimental region. LST estimation from the images acquired at the first two days was evaluated by MODIS LST products. The results showed that the combination of the VIRR 4th and 5th bands obtained higher LST estimation accuracy, compared with the combination of the VIRR 4th and MERSI 5th band. In addition, it was found that the estimated LST from VIRR 4 and 5 band images had a systematic bias (-1.664) compared with MODIS LST. The systematic bias was used to revise the LST estimation model. The revised model was validated by the images acquired at February 11. The results showed that the correlation coefficient between the estimated LST and MODIS LST was 0.877, and the RMSE was 1.33K. Compared with FY3 LST products, the estimated LST from the model was comparable to MODIS LST product.