In order to improve the recognition effect of structure damage, an effective damage evaluation method, based on stiffness variation index (SVI) and least squares support vector machine optimized by improved particle swarm optimization algorithm, is proposed. The method adopts a finite element model modified by modal testing to calculate SVI. Firstly, the SVI data are investigated to detect damage locations. Secondly, these SVI data serve as the feature vectors to be input to least squares support vector machine for calculating the hyperparameters and optimization model of the structure damage evaluation is established so as to predict the size of damage. Finally, a case study on a crane girder damage evaluation demonstrates that the method can accurately determine the actual conditions of the girder structure, and it enhances identification accuracy and convergence rate. It is an effective health evaluation method