For conventional power systems, the forced outage of components is the major cause of load shedding. Unreliability tracing is utilized to allocate the total system load-shedding risk among individual components in accordance with their different contributions. Therefore, critical components are identified and pertinent measures can be taken to improve system reliability. The integration of wind power introduces additional risk factors into power systems, causing previous unreliability tracing methods to become inapplicable. In this paper, a novel unreliability tracing method is proposed that considers both aleatory and epistemic uncertainties in wind power output and their impacts on power system load-shedding risk. First, modelling methods for wind power output considering aleatory and epistemic uncertainties and component outages are proposed. Then, a variance-based index is proposed to measure the contributions of individual risk factors to the system load-shedding risk. Finally, a novel unreliability tracing framework is developed to identify the critical factors that affect power system reliability. Case studies verify the ability of the proposed method to accurately allocate load-shedding risk to individual risk factors, thus providing decision support for reliability enhancement.