The authors measured truncation artifacts using a phantom model of the human thorax. Reconstructions were performed with both iterative cone-beam transmission maximum likelihood (TRML) and cone-beam filtered backprojection (CB-FBP). When CB-FBP was used, image truncation created 'streak' artifacts of increased density, distorted the elliptical phantom body contour into a circle with reduced radius, and increased the apparent density of the lungs. SPECT attenuation compensation using this truncated CB-FBP map produced line-source activities that were only 65% of the correct 'in-air' values. Thus, CB-FBP attenuation maps will generally be unacceptable for clinical SPECT attenuation compensation. The truncation-streak artifacts were eliminated with TRML, but in some regions the body contour was not sharply defined, and the reconstructed density was low in the lungs and nonzero outside the body contour. An elliptical support prior eliminated the 'in-air' density and improved the accuracy in the lungs, as determined by narrow-beam transmission measurements with a germanium detector. For iterations above twenty-four, TRML images were noisier than CB-FBP images with a Hann filter, and at least thirty iterations were required for accurate reconstruction. SPECT line-source attenuation compensation with a truncated TRML map was accurate to within 20%, depending upon the slice number and the source location.