Compare to the core sampling method for obtaining pavement layer parameters, which is time consuming, labor intensive, and provides limited coverage, GPR method is far more economical, faster, and easier to use. However, one defect of the GPR method is that the measurement tolerance is, so far, lower than the core sampling method. To improve the measuring accuracy of GPR method is a very important task to the GPR developers. For the rebar-free pavement, many algorithms have been proposed to accurately determine the layer parameters from GPR data. For the rebar-reinforced concrete pavement, the rebar's effect is an important factor that influences the determination of pavement thickness from GPR data. Though some features of rebar's reflection have been studied, like the data collected along traverses showing rebar as individual scattering hyperbola, or rebar responding strongly to the incident GPR waves only when the rebar is located parallel to the polarization of GPR antennas, how to extract rebar's scattering fields from the measured GPR data and cancel rebar's influence in pavement thickness estimation have not been reported. In this paper, a new algorithm for extracting rebar's reflection fields from measured GPR data is proposed. This algorithm uses two-dimensional GPR data to establish a set of functional, the variables of which include rebar depths, rebar spacing, pavement thickness, pavement electrical parameters, GPR transmitted waveform and rebar's reflection fields. By minimizing the established functional, the above variables can be determined. The proposed algorithm not only extracts the rebar's reflection fields, but also solves the other pavement parameters. With the implementation of the proposed algorithm, the measured results on highways show that the measurement tolerance is less than 5%.