This paper investigates high-resolution reconstructions from efficiently sa
mpled data in parallel-beam tomography, in particular, local tomography. A
class of sampling schemes is defined and characterized, and it is shown tha
t the standard scheme and the interlaced scheme of Cormack (Cormack A hi 19
78 Phys. Med. Biol, 23 1141-8) are most promising. An error analysis for th
e filtered backprojection algorithm for both global and local tomography is
presented. The analysis provides insights on how to realize the theoretica
lly superior resolution of the interlaced scheme in practice. A numerical e
xperiment with real data indicates the feasibility of high-resolution local
tomography using the interlaced scheme.