Due to the low accuracy rate of maximum temperature prediction in Guizhou Province, the refined guidance forecast SCMOC data and meteorological observations data from 2013 to 2018 were used to study maximum temperature forecast for 24-72 hours of 85 stations in Guizhou. The horizontal prediction model (F1), the vertical forecast model (F2) and the combination model with F1 and F2 (Fzh) were established to forecast daily maximum temperature in Guizhou. The results show that Fzh took a best performance among three models, and for these three models, both mean root-mean-square error (RMSE) and accuracy rate were improved to varying degrees. Compared with SCMOC, the forecast results of F1 in spring and summer were better than that in autumn and winter, and the improvement in the northern areas of Guizhou Province was more obvious than southern areas. The forecast results of F2 were improved all the year round, and the overall improvement was relatively stable. The forecast accuracy rate of Fzh was obviously better than F1 and F2, and RMSE decreased significantly, the mean RMSE decreased 1.0 to 2.0 ℃, the accuracy rate increased by 11% to 13%.