The dosimetric accuracy of treatment planning systems (TPSs) decreases for locations outside the treatment field borders. However, the true accuracy of specific TPSs for locations beyond the treatment field borders is not well documented. Our objective was to quantify the accuracy of out-of-field dose predicted by the commercially available Eclipse version 8.6 TPS (Varian Medical Systems, Palo Alto, CA) for a clinical treatment delivered on a Varian Clinac 2100. We calculated (in the TPS) and determined (with thermoluminescent dosimeters) doses at a total of 238 points of measurement (with distance from the field edge ranging from 3.75 to 11.25 cm). Our comparisons determined that the Eclipse TPS underestimated out-of-field doses by an average of 40% over the range of distances examined. As the distance from the treatment field increased, the TPS underestimated the dose with increasing magnitude--up to 55% at 11.25 cm from the treatment field border. These data confirm that accuracy beyond the treatment border is inadequate, and out-of-field data from TPSs should be used only with a clear understanding of this limitation. Studies that require accurate out-of-field dose should use other dose reconstruction methods, such as direct measurements or Monte Carlo calculations.