Abstract: The paper aims to test the accuracy of land surface temperature retrieval based on FY-3/VIRR data. The land surface temperature and vegetation coverage of Shandong in spring, summer, autumn and winter (April, July, October and January) were retrieved by means of the improved Becker and Li split window algorithm and pixel dichotomy, using FY-3/VIRR data. The authors analyzed the correlations between land surface temperatures (T_LS) retrieved from FY-3/VIRR data and synchronized air temperature (T_a) and 0 cm soil temperature (T_S) at 120 weather stations of Shandong Meteorological Bureau. The results showed that: the land surface temperature of Shandong Province retrieved from FY-3/VIRR data was consistent with the actual ground condition and could reflect the land surface temperature distribution intuitively. Meanwhile the T_LS had a significantly negative correlation with vegetation coverage. But for inhomogeneous land surface, the correlation coefficients between T_LS and T_a and T_S were unstable, and the correlation coefficient of different seasons had a great difference. But there was a stable linear positive correlation between (T_S-T_a) and (T_S-T_LS), and the determination coefficients R² of different seasons were greater than 0.52.