Accidents occur frequently while constructing deep foundation pits for metro stations, thereby risking substantial economic losses and casualties. To subject the construction of such pits to scientific and rational risk assessment and overcome the limitations of existing risk evaluation models and risk fusion problems, proposed here is a risk-assessment model for such pits based on fuzzy evidential reasoning and the two-tuple linguistic analytic network process (TL-ANP). First, the risk loss indicators are optimized, the weights of different risk events and of each risk loss indicator in the metro-station deep-foundation-pit construction project are calculated using TL-ANP, and trapezoidal fuzzy numbers are used to describe the occurrence probability of each risk event and loss. Second, relying on a table of expert weight indices, the best–worst method based on generalized interval-valued trapezoidal fuzzy numbers is used to determine the weights of experts. Finally, the overall risk grade of the construction project is evaluated by aggregating the risk levels of all risk events through an evidence-reasoning algorithm. The analysis results for a deep foundation pit for a station on Line 5 of Nanning Metro show that the model provides a quantitative basis for determining expert weights and risk loss weights reasonably and improving the reliability of the evaluation system. Also, not only does applying the method show that such a construction project can be judged as having a certain risk grade, but more importantly it can identify the key factors and loss indicators affecting the overall risk grade of the pit, whereupon risk control measures can be adopted in a targeted manner. In comparison with traditional methods, the proposed method is shown to be practical and effective, providing a reference basis for analyzing the risks of similar projects in the future and guaranteeing construction safety.