Traffic noise exposures were measured at various locations adjacent to an Interstate highway and compared with sound levels predicted by the Federal Highway Administration Traffic Noise Model (TNM). The prediction procedure underestimated the measured sound attenuation by 6 to 12 A-weighted decibels. Various TNM site model configurations were evaluated in an effort to improve agreement between measurements and predictions. For the site tested--a severe case with relatively distant receptors and extreme topography--variations in ground impedance (including a median ground zone) had little benefit or were counterproductive, while adding topographic detail via terrain lines helped somewhat. The best agreement resulted from the incorporation of a tree zone for the wooded site. However, this benefit is thought to be chance, because the site was not only relatively lightly wooded but also thinly foliaged at the time of the on-site measurements.