Modal analysis technique is used in order to characterize the human head-neck system in vivo. The extracted modal characteristics consist of a first natural frequency at 1.5 Hz associated to neck extension and a second mode at 6 Hz associated with head translation or neck retraction. By recording experimentally the apparent mass of dummies head-neck system under the same experimental condition as the volunteer subjects, it was possible to compare the human and the dummies frequency response functions and to evaluate their bio-fidelity. The evaluation methodology based on validation parameters extracted in the frequency domain is firstly tested on frontal and side impact dummies, Hybrid III and Eurosid. It was pointed out through their first natural frequency at around 6 Hz that this dummies present much too high rigidity for the extension mode and no retraction mode at all at higher frequencies. Frequency response analysis in terms of apparent mass was then performed on three rear impact dummies, the Hybrid III + TRID-neck, the BioRID II and RID2 v0.0. TRID showed a slightly improved extension behavior with a first natural frequency at 4.5 Hz, but not yet a retraction mode. Further improvements were detected with the proposed methodology for BioRID II and RID2 v0.0 which presented very similar behaviors characterized by a more flexible neck extension (first natural frequency around 2.5-3 Hz against 1.4 Hz in vivo) and the introduction of the retraction mode. This second mode however is set at a second natural frequency of 10 Hz for both dummies against 6 Hz recorded in vivo, illustrating a much too rigid head retraction motion. Beside dummy evaluation this study also gives new insight into injury mechanisms given that a given natural frequency can be related to a specific neck deformation.