We studied the performance of linear scatter correction methods for si
ngle-photon imaging with Tc-99m and Tl-201, using a numerical model of
the Rollo phantom and measurements with a gamma camera modified to re
cord position and energy information in list mode form. We compared th
e performance of these methods to per-image optimized linear methods a
nd to locally adaptive linear methods, and developed estimates of the
limits on accuracy of scatter correction imposed by the presence of Po
isson noise. For both Tc-99m and Tl-201 imaging at a fixed depth, part
icularly at low count rates, the performance of dual-window methods, o
r of adaptive methods, is near the best possible for linear methods. S
moothing of the scatter estimate results in minor improvement for Tl-2
01. Substantial gaps between the performance of any of these linear me
thods and the limits imposed by Poisson noise remain and are due prima
rily to bias, with the gap for Tl-201 being larger than that for Tc-99
m.