The compaction of asphalt concrete significantly affects long-term pavement performance. Although coring provides a relatively accurate way of assessing in-place density at specific locations, the coverage of the assessment is limited, especially at longitudinal joint locations. This can be particularly problematic because it is difficult to identify problematic locations that are likely to fail prematurely using current compaction assessment methods. Ground penetrating radar (GPR) provides an attractive nondestructive testing alternative for evaluation of compaction quality, especially with recent significant improvements in the GPR technology for this specific application. However, assessment of the air void content of the asphalt mix from the GPR-measured dielectric constant of the surface requires conversion of dielectric variation to air void content variation, which is the subject of this paper. An alternative to the commonly used model is proposed, leading to more justifiable predictions for low values of dielectric constants. The proposed model was used to interpret data from a 7-mi long asphalt overlay construction project. The results of the interpretation as compared with the results obtained with the conventional model show an improvement on the stability of the prediction at low air void contents, especially when core calibration data are limited and uncertainty is considered. These results are promising in the direction of reducing field cores necessary to have a stable model providing continuous compaction assessment of new asphalt pavement construction.