In this paper, a new approach to structural damage localization is presented using as damage feature the interpolation error related to the use of a spline function in modeling the operational deformed shapes of the structure. Statistically significant variations of the interpolation error between the undamaged and the inspection phase indicate the onset of damage. A threshold value of the damage feature is defined in terms of the tolerable probability of false alarm to select variations of the interpolation error because of damage from those due to random sources. The method is successfully applied to a calibrated model of the factor building a real densely instrumented building at the University of California, Los Angeles. Results show that the method is effective for damage localization for both single and multiple locations of damage also in case of responses corrupted by noise. Copyright © 2014 John Wiley & Sons, Ltd.