Nonlinearity is one of important frontier issues in the field of modern hydrology. It is of great significance to discover the variation and the complexity of the hydrological system. On the basis of hydrological and meteorological data in the headwater region of the Kaidu River from 1960 to 2012,multi- scale characteristics of runoff variability were analyzed using the ensemble empirical mode decomposition method to investigate nonlinear change characteristics of runoff at different time scales and multi-scale responses to climate fluctuation. We found that in the past 50 years,the overall runoff of Kaidu River in Xinjiang has exhibited a significant non- linear increasing trend,and changes have manifested at quasi-2~3 and quasi-4~5 year inter-annual scales and showed quasi-10~11 and quasi-26~27 year inter-decadal scales. Variance contribution rates of each periodic component showed that inter- annual change held a dominant position in the longterm change of runoff,and inter-decadal change also played an important role in the overall runoff change for Kaidu River. Reconstructed inter- annual variation describes the fluctuation state of original runoff during the study period,that is to say the trends of both are basically the same. The reconstructed inter-decadal variability shows that the runoff of Kaidu River underwent a significant transition in 1995 from low to high,namely the state of runoff anomaly shifting from a negative phase to a positive abundance water period. In addition,we found that the correlation between runoff and precipitation is more relevant and significant at both the inter-annual and inter-decadal scales compared to the correlation between runoff and temperature and potential evaporation, indicating that precipitation is the decisive factor for changes in runoff. Corrections between runoff and precipitation,temperature and potential evaporation are more significant and relevant at the inter-decadal scale than inter-annual scale,suggesting that the inter-decadal scale is more suitable for investigating responses of runoff dynamics to climate fluctuation.