NMC method is widely used to statistically analyze background error covariances currently.In this paper, based on a month-long series of model forecast fields in August 2002 and February 2003, the background error covariance of high-resolution mesoscale model MM5V3 in cold and warm seasons, which is about to pre-operationally run in Beijing region, is computed using NMC method.The detail climatological statistical characteristics are present as well.The results show that different model variables has different background error characteristics, even the same model variable has different background error characteristics when model resolution or model vertical-level changes.The characteristics are also different in cold and warm seasons, mainly wind field.All those characteristics are corresponding to averaged weather condition in the forecast region.Consequently, 3DVAR should be done respectively for different model domains.In practical run of MM5 3DVAR system in Beijing, method used to compute and update background error covariance should be developed incorporating actual seasonal variation of weather condition.