The objectives of this study are (1) to develop an accurate and efficient fatigue analysis procedure that can be used in reliability analysis and reliability-based design optimization (RBDO) of composite wind turbine blades; (2) to develop a wind load uncertainty model that provides realistic uncertain wind load for the reliability analysis and the RBDO process; and (3) to obtain an optimal composite wind turbine blade that satisfies target reliability for durability under the uncertain wind load. The current research effort involves: (1) developing an aerodynamic analysis method that can effectively calculate detailed wind pressure on the blade surface for stress analysis; (2) developing a fatigue failure criterion that can cope with non-proportional multi-axial stress states in composite wind turbine blades; (3) developing a wind load uncertainty model that represents realistic uncertain wind load for fatigue reliability of wind turbine systems; (4) applying the wind load uncertainty model into a composite wind turbine blade and obtaining an RBDO optimum design that satisfies a target probability of failure for a lifespan of 20 years under wind load uncertainty. In blade fatigue analysis, resultant aerodynamic forces are usually applied at the aerodynamic centers of the airfoils of a blade to calculate stress/strain. However, in reality the wind pressures are applied on the blade surface. A wind turbine blade is often treated as a typical beam-like structure for which fatigue life calculations are limited in the edge-wise and/or flap-wise direction(s). Using the beam-like structure, existing fatigue analysis methods for composite wind turbine blades cannot cope with the non-proportional multi-axial stress states that are endured by wind turbine blades during operation. Therefore, it is desirable to develop a fatigue analysis procedure that utilizes detailed wind pressures as wind loads and considers non-proportional multi-axial stress states in fatigue damage calculation. In this study, a 10-minute wind field realization, determined by a 10-minute mean wind speed V10 and a 10-minute turbulence intensity I10, is first simulated using Veers’ method. The simulated wind field is used for aerodynamic analysis. An aerodynamic analysis method, which could efficiently generate detailed quasi-physical blade surface pressures, has been developed. The generated pressures are then applied on a high-fidelity 3-D finite element blade model for stress and fatigue analysis. The fatigue damage calculation considers the non-proportional multi-axial complex stress states. A detailed fatigue damage contour, which indicates the fatigue failure locally, can be obtained using the developed fatigue analysis procedure. As the 10-minute fatigue analysis procedure is deterministic in this study, the calculated 10-minute fatigue damage is determined by V10 and I10. It is necessary to clarify that the rotational speed of the wind turbine blade is assumed to be constant (12.1 rpm) and the pitch angle is fixed to be 0 degree for different wind conditions, since the rotational speed control and pitch angle control have not been considered in this study. For predicting the fatigue life of a wind turbine, a fixed Weibull distribution is widely used to determine the percentage of time the wind turbine experiences different mean wind speeds during its life-cycle. Meanwhile, fixed turbulence intensities are often used based on the designed wind turbine types. These simplifications, i.e., fixed Weibull distribution and fixed turbulence intensities, ignore the realistic uncertain wind load when designing a reliable wind turbine system. In the real world, both the mean wind speed and turbulence intensity vary constantly over one year, and their annual distributions are different at different locations and in different years. Thus, it is necessary to develop a wind load uncertainty model that can provide a realistic uncertain wind load for designing reliable wind turbine systems. In this study, 249 groups of measured wind data, collected at different locations and in different years, are used to develop a dynamic wind load uncertainty model. The dynamic wind load uncertainty model consists of annual wind load variation and wind load variation in a large spatiotemporal range, i.e., at different locations and in different years. The annual wind load variation is represented by the joint probability density function of V10 and I10. The wind load variation in a large spatiotemporal range is represented by the probability density functions of five parameters, C, k, a, b, and τ, which determine the joint probability density function of V10 and I10. In order to obtain the RBDO optimum design efficiently, a deterministic design optimization (DDO) procedure of a composite wind turbine blade has been first carried out using averaged percentage of time (probability) for each wind condition. A wind condition is specified by two terms: 10-minute mean wind speed and 10-minute turbulence intensity. In this research, a probability table, which consists of averaged probabilities corresponding to different wind conditions, is referred as a mean wind load. The mean wind load is generated using the dynamic wind load uncertainty model. During the DDO process, the laminate thickness design variables are tailored to minimize the total cost of composite materials while satisfying the target fatigue lifespan of 20 years. It is found that, under the mean wind load condition, the fatigue life of the initial design is only 0.0004 year. After the DDO process, even though the cost at the DDO optimum design is increased by 31.5% compared to that at the initial design, the predicted fatigue life at the DDO optimum design is significantly increased to 19.9995 years. Reliability analyses of the initial design and the DDO optimum design have been carried out using the wind load uncertainty model and Monte Carlo simulation. The reliability analysis results show that the DDO procedure reduces the probability of failure from 100% at the initial design to 49.9% at the DDO optimum design considering only wind load uncertainty. In order to satisfy the target 2.275% probability of failure, it is necessary to further improve the fatigue reliability of the composite wind turbine blade by RBDO. Reliability-based design optimization of the composite wind turbine blade has been carried out starting at the DDO optimum design. Fatigue hotspots for RBDO are identified among the laminate section points, which are selected from the DDO optimum design. Local surrogate models for 10-minute fatigue damage have been created at the selected hotspots. Using the local surrogate models, both the wind load uncertainty and manufacturing variability has been included in the RBDO process. It is found that the probability of failure is 50.06% at the RBDO initial design (DDO optimum design) considering both wind load uncertainty and manufacturing variability. During the RBDO process, the normalized laminate thickness design variables are tailored to minimize the total cost of composite materials while satisfying the target 2.275% probability of failure. The obtained RBDO optimum design reduces the probability of failure from 50.06% at the DDO optimum design to 2.28%, while increasing the cost by 3.01%.