The purpose of this study was to determine the accuracy of five commonly used intensity-modulated radiation therapy (IMRT) treatment planning systems (TPSs), 3 using convolution superposition algorithms or the analytical anisotropic algorithm (CSA/AAAs) and 2 using pencil beam algorithms (PBAs), in calculating the absorbed dose within a low-density, heterogeneous region when compared with measurements made in an anthropomorphic thorax phantom. The dose predicted in the target center met the test criteria (5% of the dose normalization point or 3 mm distance to agreement) for all TPSs tested; however, at the tumor-lung interface and at the peripheral lung in the vicinity of the tumor, the CSA/AAAs performed better than the PBAs (85% and 50%, respectively, of pixels meeting the 5%/3-mm test criteria), and thus should be used to determine dose in heterogeneous regions.