The present work aims at applying and comparing some methods for vibration-based damage detection of civil structures starting from ambient vibration data. The analyzed procedures are based on a dynamic identification of the modal parameters that is performed in output-only conditions; this is a typical situation that occurs when structural health monitoring strategies are applied to civil structures. The here investigated damage-sensitive features are the modal parameters, the modal flexibility matrix and the damage-induced deflection, due to unitary inspection loads, of the identified structure; possible variations in these parameters can be adopted in order to detect, localize and quantify the damage. A damaged condition is analytically simulated through a stiffness reduction in the elements of a RC shear-type plane frame. The output-only modal identification is performed through the Eigensystem Realization Algorithm, combined with the Random Decrement technique, on the simulated responses due to a white noise ground motion of both the undamaged and the damaged structures. At the end, the effectiveness of the different identified damage features is evaluated and the accuracy related to the identified modifications, due to damage, is determined through a comparison with the initially assumed variations in the structural model.