Our long-term goal is to develop an ultrasonic method to characterize the thickness, stiffness and porosity of the cortical shell of the femoral neck, which could enhance hip fracture risk prediction. To this purpose, we proposed to adapt a technique based on the measurement of guided waves. We previously evidenced the feasibility of measuring circumferential guided waves in a bone-mimicking phantom of a circular cross-section of even thickness. The goal of this study is to investigate the impact of the complex geometry of the femoral neck on the measurement of guided waves. Two phantoms of an elliptical cross-section and one phantom of a realistic cross-section were investigated. A 128-element array was used to record the inter-element response matrix of these waveguides. This experiment was simulated using a custom-made hybrid code. The response matrices were analyzed using a technique based on the physics of wave propagation. This method yields portions of dispersion curves of the waveguides which were compared to reference dispersion curves. For the elliptical phantoms, three portions of dispersion curves were determined with a good agreement between experiment, simulation and theory. The method was thus validated. The characteristic dimensions of the shell were found to influence the identification of the circumferential wave signals. The method was then applied to the signals backscattered by the superior half of constant thickness of the realistic phantom. A cut-off frequency and some portions of modes were measured, with a good agreement with the theoretical curves of a plate waveguide. We also observed that the method cannot be applied directly to the signals backscattered by the lower half of varying thicknesses of the phantom. The proposed approach could then be considered to evaluate the properties of the superior part of the femoral neck, which is known to be a clinically relevant site.