Transmission computed tomography provides information needed for nonuniform attenuation correction of cardiac single photon emission computed tomography (SPECT). Nonuniform attenuation correction is accomplished using an iterative ML-EM algorithm and a projection-backprojection operation that incorporates attenuation factors measured from the reconstructed transmission map. The precision and accuracy of the attenuation corrected emission reconstruction is a function of emission and transmission statistics. This paper presents an error propagation analysis that uses a mathematical cardiac chest phantom to simulate various combinations of total emission counts C and transmission flux I0 under ideal imaging conditions (without geometric response distortion and without scatter). The spatial average, spatial variance, and accuracy measures for a 4 x 4 pixel region in the heart are tabulated after 30 iterations of the ML-EM algorithm. The confidence intervals for these measures were determined from 1000 realizations of reconstructions from projections randomly generated with the same transmission and emission statistics. It can be shown empirically from the simulation results that the spatial %rms uncertainty for the simulated cardiac region has a simple expression: %rms2 = K1/C+K2/I0(2)+B2 where K1 and K2 are least-square estimates based on the simulation results, and B is the measured spatial %rms uncertainty for the simulation at infinite statistics. For a transmission incident flux of 1500 events per projection bin of 0.712 cm and typical clinical emission events totaling 1 x 10(5), the spatial %rms uncertainty is approximately 14%. At clinical transmission and emission statistics, the statistical noise in the simulated attenuation-corrected reconstructions are dominated by the emission statistics.