Abstract: There were 26 days of high temperature when the daily maximum air temperature (MAT) was above 35 ℃ in Shaanxi Province in July 2017, and out of the 26 days there were 14 days with the MAT being even above 40 ℃.  The continuous hot weather in this region in the month occurred twice during the periods from July 7th to 14th and from July 17th to 27th respectively. Based on hourly observation of MAT at 99 stations in Shaanxi Province reported by CMA (Chinese Meteorological Administration) and fixed time forecast of ECMWF (European Center for Medium Weather Forecasting) high resolution model, the performance of ECMWF model for the hot weather forecast and temperature correction by linear regression method were evaluated. The evaluation methods included average error, average absolute error and accuracy with absolute error less than 1 ℃ and 2 ℃. The results showed an obvious regional difference in forecast ability of the model, whose accuracy of the MAT is higher in Guan Zhong and lower in Southern Shaanxi. The regional continuous hot weather could have been better forecasted by the model during the forecast time of 144 hours, but the number of hot weather days predicted was less than the observation days in most areas of Shaanxi province, and it failed to forecast the hot weather days in most areas of southern Shaanxi. This was because that the values of MAT from prediction were less than the observation in most areas of Shaanxi, and the MAE (mean absolute error) in southern Shaanxi was more than that in other areas of Shaanxi from 14:00 to 17:00 BST. With the extension of forecast hours, the capability of the linear regression method to correct air temperature was decreased in wave mode. The linear regression method had positive effect to correct air temperature during the forecast time of 168 hours, which is better between 14:00 PM and 01:00 AM the next day than that in any other times, and there was a significant negative correlation relationship between the correction ability of the method and forecast ability of the model. The linear regression method could increase the accuracy and decrease the MAE of MAT. The ME (mean error) of MAT was closer to zero, and the hot weather forecast with a daily MAT being above 35 ℃ or 40 ℃ was closer to the observation after correction. The research provides a rational reference for hot weather forecasts from ECMWF model in summer, and it is noted further studies during other time periods are needed to verify the universality of the conclusion as this study was focused on the period of July 2017.