The Method for ObjectBased Diagnostic Evaluation (MODE) has been widely applied in spatial evaluation in recent years. MODE is affected by many parameters such as precipitation critical value, convolution radius and attribute weight; the application effect of MODE depends on reasonable selection of convolution radius and accurate characterisation of spatial similarity between forecast and observation fields. Taking the CMPA as observation, MODE and FSS (Fractions Skill Score) are used to test the CMAGD 24 h daily precipitation forecast based on 54 precipitation cases in Guizhou in this paper. The number of objects extracted by MODE falls with convolution radius; too small convolution radius easily results in too many precipitation objects extracted; if the convolution radius is too large, local precipitation information will be lost and precipitation objects cannot be extracted from the precipitation field. Therefore, an appropriate convolution radius should be adopted to extract precipitation objects with MODE. It is found that the MMI (the Median Maximum Interest Value) of MODE is very sensitive to the convolution radius change and even has a mutation, so it cannot stably represent the overall spatial similarity of precipitation fields. Based on the MMI, the area weight is introduced to construct the AMMI (the Area Mean of Maximum Interest Value) to distinguish the contribution of different objects. The AMMI is more reasonable to characterise the overall similarity of the forecast and observation precipitation fields, and is unaffected by the number of precipitation objects, which is more stable than the MMI. In general, AMMI is larger than FSS, and the difference in the change of AMMI and FSS with spatial scale is due to the different calculation basis. According to the change of object’s total area with the convolution radius, precipitation can be divided into largearea precipitation and local precipitation. The average total area of largescale precipitation grows with the convolution radius, while AMMI has little change. As the convolution radius goes up, the average total area and AMMI of local precipitation go down. Taking the maximum convolution radius which makes the total area change not exceeding 10% in the observation field as the critical radius, there is a large difference between the probability of the critical radius of largescale precipitation and local precipitation from 0.05° to 0.4°. Local precipitation is sensitive to the selection of convolution radius and determining the convolution radius with critical radius is helpful to retain most of the information of the precipitation field.