The evaluation of fatigue damage accumulation on wind turbine support structures under operational conditions is heavily influenced by a number of uncertainties. These uncertainties may, firstly, be attributed to the highly variable and complex environmental loads, and secondly, to the unavoidable modelling errors which mainly originate from the inherent randomness in both material properties and fatigue resistance of structural components. It is therefore essential that assessment of fatigue life is carried out within a probabilistic framework; one that accounts for the stochastic nature of the phenomenon. The present study proposes a strategy for real-time reliability prediction of accumulated fatigue damage on wind turbine support structures by taking into account the above-mentioned uncertainties. To this end, the availability of structural monitoring information for the identification of the global response on wind-turbine support structures is exploited in order to address the discrepancies between actual and predicted damage accumulation. This is carried through utilization of an augmented version of the Kalman filter, which is capable of jointly estimating the response and the unknown inputs of the structure while relying on a limited number of noisy observations and a presumably uncertain model of the real system. A fixed-lag smoother is further deployed for the attenuation of the estimation error in an on-line mode and the smoothed stochastic estimates of the response are propagated over the model at the level of stresses. The accumulated damage along with the corresponding reliability level is finally predicted using a stochastic nonstationary fatigue damage model. The proposed scheme is demonstrated via implementation on the NREL 5.0 MW wind turbine under different operational conditions, on the basis of dummy vibration data generated via the FAST software. @en