Considering the premise of consistency hypothesis is no longer satisfied for the traditional runoff frequency analysis methods, so the Generalized Additive Model of Location, Scale and Shape (GAMLSS) was employed to detect the non-stationarity of the annual runoff series (1956-2014) in Manas River and correlation analysis method was used to build time varying moment models based on time and climatic factors as covariate variables respectively, whose effects were compared with the traditional model (P-III). The results show that: NAO index could be as one of the climatic factors because of its best relativity with annual runoff serials and the correlation coefficient is-0.322; the optimal distribution model which is based on the accumulated temperature deficit, precipitation index and NAO as covariate variables effectively describes dynamic characteristics of annual runoff in Manas River under climate change. Under different design guarantee rates, annual design runoff by the LOGNO model is larger than the traditional model (P-III Model0) by 3.08%-16.10%. Particularly, the monthly runoff design value is rather different from Model0. The research results provide reference for the efficient use of resources and scientific management of Manas River.