The accuracy of dose calculations from a pencil-beam algorithm developed specifically for arc electron beam therapy was evaluated at 10 and 15 MeV. Mid-arc depth-doses were measured for 0 degrees and 90 degrees arcs using 12 and 15 cm radius cylindrical water phantoms. Calculated depth-doses for the 90 degrees arced beams in the build-up region were as much as 3% less than measured values; the maximum dose was similar in magnitude but at a greater depth; and the therapeutic depth, R80, was 2-4 mm deeper. Calculated values of output (dose per monitor unit) at the depth of the maximum calculated dose were compared with measured values; for arcs ranging from 0-90 degrees, 12 and 15 cm radius water phantoms, and collimator widths of 4, 5 and 6 cm, results showed differences as great as 7%. Isodose countours for a 90 degrees arc were also measured in a 15 cm radius PMMA phantom. At the depth of maximum dose the algorithm predicted doses in the penumbral regions, both with and without collimation, which agreed within a few per cent of measured values. The largest discrepancies were 5%, which occurred in the penumbral portion of the depth-dose fall-off region. Differences between measurement and calculation are not believed to be clinically significant and are believed to be primarily due to the fact that the algorithm models neither large-angle scattering nor the effects of range straggling on the pencil-beam dose distribution.