Climate change and human activities often result in non-stationarity in hydrological series, but no general consensus on the methods for calculation of the return period of non-stationary series has been reached yet. In this study, we compared several methods, including traditional method, time point method, expected waiting time (EWT), and component analysis, and took a case study of the annual runoff series for the Linjiacun station on the Wei River. A time-varying parameter model was used to fit the probability distribution of non-stationary series and calculate the design values of runoff at different return periods. The results indicate that the traditional method fails to correctly describe the probability distribution of hydrological series featured with trend alteration. In the condition of a decreasing annual runoff in the study area, the probability of small runoffs is increasing year by year, while the runoffs corresponding to long return periods does not show a decreasing trend. In comparison of EWT and time point method, the former can take the influence of trend alteration into account, while the latter has the advantage of easier calculation. The analysis and comparison of the methods presented in the paper would lay a basis for analysis of non-stationary hydrological series.