Abstract: A method of truss damage identification is proposed based on the principal component analysis and multivariate control chart using frequency response functions as basic variables. Original data matrix is formed using frequency response function obtained from truss dynamic test. With principal component analysis, the variable space dimensions of original data can be reduced. Then the characteristics of original data can be analyzed and extracted from lower dimension space using first several principal components which contain the most original data information. The truss damages can be identified by separating the singular data in the control chart, which is formed using first several order principal components. In order to validate the reliability of the proposed method, a full-size truss was tested in four kinds of damage cases concerning single and double element damages. The result shows that the proposed method is reliable and has a good adaptability for slight non-linearity truss under the noise environment. Since the proposed method does not need modal parameters and avoids the errors of modal extraction, it is straightforward and accurate for truss damage identification.