Owing to the climate change and human activities, it is important to use non-stationary methods for design flood calculation. Taking Longtanzhai hydrological station on the Panlong River in Yunnan Province as a case, using the observed, simulated, and predicted precipitation and flood datasets based on the GAMLSS model and Equal Reliability method, the stationary and non-stationary design floods were calculated. The results show that, (1) the non-stationary probability models with precipitation as the covariate are better than the non-stationary probability models with time as the covariate and the stationary probability model. The precipitation-covariate models can both simulate the trend and fluctuation of floods. (2) The stationary design flood (269-423 m3/s) is generally greater than the non-stationary one (182-432 m3/s) at Longtanzhai hydrological station. (3) It is more reasonable to use precipitation as a covariate for the study case. The results could provide the foundation for flood management as well as the design of water conservation projects.