The impact of uneven precipitation on the runoff amount is very significant in the Yellow River basin. The years of 1983 and 1985 are both rich hydrological years with precipitation of 501 mm and 502 mm, respectively, but the natural runoff amounts of the two years at the Huayuankou station are 76 5 and 63 8 billion m\+3. This means that there is a 12 7 billions m\+3 difference of runoff amount when the rainfalls are almost the same. This paper analyzes the impact of precipitation unhomogeneity on runoff generation. Firstly, the impact of spatial unhomogeneity is analyzed. The normal method is to divide the Yellow river basin into five sections , namely, upper Lanzhou section, Lanzhou-Hekou section, Hekou-Longmen section, Longmen-Sanmenxia section, Sanmenxia-Huayuankou section. When we analyze by using this division, the simulation result in the four sections of lower Lanzhou region is quite good, but the simulation result in upper Lanzhou section is bad. Concretely, the simulation result is that the runoff amount in 1983 is 6 7 billion m\+3 more than that of 1985 in the middle reaches, which is very close to the measured 5 4 billion m\+3. And the result also shows that the reason of the difference of the runoff amount generated in the middle reach is that the precipitation and runoff in the Longmen-Sanmenxia section is different. In the Lanzhou-Hekou section, the difference of precipitation doesn’t impact the runoff very significantly since the precipitation-runoff generation coefficient is very small. The runoff amount generated in this section is very small. The precipitation and runoff in the Hekou-Longmen section and Sanmenxia-Huayuankou section are different but total amounts are very close. Therefore the reason for the difference of water amount is the runoff generation in the upper Lanzhou section and Longmen-Sanmenxia section. According to simulation result through the precipitation-runoff relation, the precipitation and runoff amount in 1983 is less than that of 1985, but the measured figure of precipitation in 1983 is less than that of 1985 while the runoff in 1983 is 7 5 billion m\+3 more than that of 1985. This means that the normal division can not explain the runoff generation characteristics. Then we make further division of the spatial scale by subdivision of upper Lanzhou section into upper Tangnaihai section and Tangnaihai-Lanzhou section. The result is that, although the precipitation of upper Lanzhou section of 1983 is less than that of 1985, the precipitation and runoff of upper Tangnaihai section of 1983 are larger than those of 1985, and the precipitation and runoff of Tangnaihai-Lanzhou section of 1983 keep less than those of 1985, while this upper Tangnaihai section is the so-called high water generation rate section. This result can explain why the runoff of upper Lanzhou section in 1983 is larger than that of 1985 while the precipitation of upper Lanzhou section in 1983 is less than that of 1985. Such a subdivision can clearly indicate that the precipitation and runoff relation in upper Tangnaihai section is dominant in the upper Lanzhou sction. Therefore spatial division must be drawn into Tangnaihai section at least, and the result could explain the precipitation-runoff relation. Secondly, we subdivide the temporal scale into flood period and non flood period. Analysis result shows that such a time scale division doesn’t impact very much of precipitation-runoff relation and that, at the verse, the simulation result of precipitation-runoff relation is worse than that in year-scale.