Based on the percentile method, a distinction method for the extreme weather events is proposed by the construction of the climatologically equally likely intervals firstly.Then, the corresponding forecasting evaluation on extreme weather events is performed for 500 hPa geopotential height over Eurasian area in summer 2003.The results are as follows:1) In the analyzed period and region, the frequency of the extreme events is slightly higher than the climatological average situation.In NCEP-EPS, the forecast frequency is easy to become higher for the lower extreme events, but relatively reliable, whereas it is decreased with the extension of lead-time for the higher extreme events.When the lead-time is less than 3 d (more than 5 d), the forecast frequency is higher(lower) with relatively less reliability.2) EV(economic value) analysis shows that the performance of NCEP-EPS probabilistic forecast for the higher(lower) extreme events is better than that for the normal weather events.3) Forecast hit rate analysis shows that the hit rate is lower in the high-probability threshold.Further analysis points out that it is perhaps related with the lower ensemble spread in some cases.With the decrease of probability threshold, the hit rate steadily increases.It means that the spread of ensemble members can make them to cover all possible region of observation value, thereby to get the better forecasting results.