Extreme events become more and more frequent under the global warming. It is very important to provide an extreme events prediction. Based on daily minimum temperature data for more than 700 observation stations over China and 1983-2010 winter hindcasts of the first generation atmospheric-ocean coupled model in Beijing Climate Center,a new prediction of extreme winter cold days(EWCD) over China is developed by using the stepwise regression statistical downscaling model(SRSDM). Results show that the correlation coefficient of EWCD between predictions using SRSDM and observations for 1983-2010 years exceed the 95% significant level in most of China. Moreover,inter-annual variability of EWCD predicted by SRSDM is well agreed with observations during 1983-2010. The realtime prediction of EWCD in 2011 /2012 using the SRSDM was carried out. It is very well that the prediction is successfully and the prediction of EWCD by SRSDM in 2011 /2012 is basically in accordance with the observation. Above all proved the method to predict EWCD over China by SRSDM can be employed in operational application.