Based on analysis of chaos theory and adaptability in runoff system, the chaotic analysis method is put forward. Monthly runoff series is reconstructed in phase space, and chaotic characteristics are identified and analyzed with Yellow River as the research object. The conclusions are as follows: natural monthly runoff is larger than the measured in saturated correlation dimension. At least 4 independent variables, up to 8 are needed to properly describe changing features of the measured monthly runoff series and build dynamic system modeling; while for natural runoff series, at least 5-6 variables， 12 at most are needed. The measured runoff series and the natural vary in chaotic characteristics in the same hydrologic station during the same period: the chaotic characteristics of downstream are stronger than upstream; the chaotic characteristics of monthly runoff from 1950s to the beginning of 21century are slightly stronger than those from 1920s to 1970s in Yellow River; The length of runoff time series has influence on the level of chaotic characteristics, the length is more longer, the level is more stronger. In a whole, the Yellow River runoff has chaotic characteristics, which provide the basis for runoff system modeling and chaotic forecasting.