Patient anatomy has complicated effects on cardiac SPECT images. We investigated reconstruction methods which substantially reduced these effects for improved image quality. A 3D MCAT (mathematical cardiac-torso) phantom which models the anatomical structures in the thorax region were used in the study. The phantom was modified to simulate variations in patient anatomy including regions of natural thinning along the myocardium, body size, diaphragmatic shape, gender, and size and shape of breasts for female. Distributions of attenuation coefficients and Tl-201 uptake in different organs in a normal patient were also simulated. Emission projection data were generated from the phantoms including effects of attenuation and detector response. We demonstrated the attenuation-induced artifacts caused by patient anatomy in the conventional FBP reconstructed images. Accurate attenuation compensation using iterative reconstruction algorithms and attenuation maps substantially reduced the image artifacts and improved quantitative accuracy. We conclude that reconstruction methods which accurately compensate for non-uniform attenuation can substantially reduce image degradation caused by variations in patient anatomy in cardiac SPECT.