Abstract: This paper presents an effective approach for system reliability analysis of megawatt wind turbine structures with the principle of maximum entropy and sample-based fractional moments. To begin with,a probabilistic model for the long-term extremevalued structural loads（e.g. the annual,the twenty-or fifty-year return period）is derived based on the order statistic theory and the three-parameter Weibull distribution. This is further used to define performance functions for the reliability analysis of wind blade and tower structures. With fractional moments estimated based on a small number of samples,probability distribution of the system performance function is recovered through the principle of maximum entropy（MaxEnt）with sample-based fractional moments（MESFM）. In numerical examples,the structural performance functions are defined by bending moments of the blade and tower,whereas the dimensionless statistics reference to material strength are derived to model the system failure events. Numerical examples have shown that the predicted structural failure probability results provided by the proposed ME-SFM approach are in closed agreement with that of the Monte-Carlo simulation. The normal and extreme turbulence models significantly influence the system failure probability level,whereas the effect needs to be carefully evaluated by various wind load cases in the IEC 61400-1 standard.