A damage detection method for uncertainty quantification using interval analysis is explored in this study. A vector auto-regression (VAR) model is established on the basis of the structural acceleration response data from the test. The Mahalanobis distance is extracted from the main diagonal of the coefficient matrix of VAR model and is adopted as the damage characteristic index. On the basis of particle swarm optimization (PSO), an interval optimization method is established, and its performance is compared with two traditional methods using a non-convex function. The damage location and damage degree are identified by interval overlap index and interval nominal value respectively. Both the numerical simulation and a laboratory frame structure test show that the presence and the severity of damage can be confidently detected even with a few data.