Seventy-six patients were studied within one week of contrast ventriculography to assess the influence of differing background assignments on election fractions derived from gated equilibrium radionuclide angiograms. 'Fixed' and 'variable' left-ventricular regions of interest and five different backgrounds were used. Inter- and intra-observer variability were determined, and 15 patients underwent a second study 2 weeks later. Variable left-ventricular regions of interest produced higher ejection fractions than fixed ones. A computer-assigned background or a ring drawn manually around the left ventricle correlated best with contrast ventriculography, giving greater serial and inter- and intra-observer variability. Automated background assignments reduced variability and increased reproducibility, though not always associated with the best correlation with contrast ventriculography. Thus a variety of left-ventricular and background regions of interest gives excellent reproducibility and accuracy.