Abstract: High resolution numerical model has a certain ability to predict the structure and evolution characteristics of the convection system, however, it is not easy to exploit the advantage of high resolution numerical model forecast using traditional verification metrics. Fuzzy verification is a recently developed and popular spatial verification method used for high resolution numerical model. It compares characteristics of the adjacent area of the corresponding point in prediction and observation fields to assess the accuracy of the forecast. Fuzzy verification considers a certain space and time uncertainty instead of completely accurate matching between prediction and observation.The Method of Fraction Skill Score, which is a kind of fuzzy verification, is used to evaluate the convective weather forecast ability in North China from 3 different high-resolution models(including Rapid Analysis and Forecast System GRAPES_Meso, GRAPES_3 km and East China Regional Numerical Model). Seven convective cases in North China with different organization modes caused by different weather systems from July to September 2017 are selected to evaluate the prediction ability of the small and medium scale convection system of high resolution models. Results show that the fraction skill score (FSS) can achieve valuable assessment information when the model prediction has a certain displacement and intensity of deviation. Another dominant advantage of FSS is that it can examine the model forecast skill scale, referring to the smallest window scale over which the forecast output contains useful information. The forecast skill scale is different from the scale of weather system, it represents the scale of spatial displacement deviation. The prediction of radar echo intensity from three models is weaker than the observation, East China Regional Numerical Model is the closest to the observation with echoes below 44 dBZ, and the intensity deviation from GRAPES_3 km is the smallest with echoes above 44 dBZ. Those performances lead to difference of FSS curve between GRAPES_3 km and East China Regional Numerical Model in the case of lower threshold and higher threshold. In order to evaluate the spatial displacement deviation of the model, the frequency (percentile) threshold is adopted, which represents the size change of the verification object. With increasing percentile threshold, GRAPES_3 km forecast skill scale basically maintains at 150-200 km, but the forecast skill scale of East China Regional Numerical Model increases gradually from 100 km to 400 km against the percentile threshold, which suggests that GRAPES_3 km model is more capable of predicting the small scale convective events.