Practitioners analyze the elevation profile of a roadway to detect localized defects and to produce the international roughness index. The prevailing method of measuring road profiles uses a specially instrumented vehicle and trained technicians, which usually leads to a high cost and an insufficient measurement frequency. The recent availability of probe data from connected vehicles provides a method that is cost-effective, continuous, and covers the entire roadway network. However, no method currently exists that can reproduce the elevation profile from multi-resolution features of the vehicle inertial response signal. This research uses the wavelet decomposition of the vehicle inertial responses and a nonlinear autoregressive artificial neural network with exogenous inputs to reconstruct the elevation profile. The vehicle inertial responses are a function of both the vehicle suspension characteristics and its speed. Therefore, the authors normalized the vehicle response models by the traveling speed and then numerically solved their inertial response equations to simulate the vehicle dynamic responses. The results demonstrate that applying the artificial neural network to the wavelet decomposed inertial response signals provides an effective estimation of the road profile.