Many different vibration-based dynamic input-output and output only data features have been used to identify structural damage and assess structural integrity. Since structural damage introduces linear or nonlinear variations into all of these features, all of them might give positive indications of damage but may not distinguish between linear or nonlinear types of damage. This information can sometimes be used to more reliably diagnose damage by first, helping to distinguish between damage, which is inherently nonlinear, and healthy nonlinearities in a baseline structure; and second, serving as an absolute damage prognosis indicator which, together with prior information about the structural mechanics, determined the degree to which a structure is damaged. A set of potential features that distinguish between linear and nonlinear damage are discussed here. These features are auto-regressive exogenous dynamic transmissiblity model coefficients in the frequency domain. The auto-regressive coefficients are used to characterize the nonlinear nature of damage states and the exogenous coefficients are used to characterize the linear nature of such states. After reviewing the theoretical development of this data model, experimental measurements from a three-story test structure are analyzed using these model coefficients and statistical features are extracted from the coefficients. By using two complementary features, a better indication of the severity of damage is obtained.