Linear polarization at the level of similar to 1% - 3% has by now been measured in several gamma-ray burst afterglows. Whereas the degree of polarization, P, was found to vary in some sources, the position angle, theta(p), was roughly constant in all cases. Until now, the polarization has been commonly attributed to synchrotron radiation from a jet with a tangled magnetic field that is viewed somewhat off-axis. However, this model predicts either a peak in P or a 90degrees change in theta(p) around the " jet break" time in the light curve, for which there has so far been no observational confirmation. We propose an alternative interpretation, wherein the polarization is attributed, at least in part, to a large-scale, ordered magnetic field in the ambient medium. The ordered component may dominate the polarization even if the total emissivity is dominated by a tangled field generated by postshock turbulence. In this picture theta(p), is roughly constant because of the uniformity of the field, whereas P varies as a result of changes in the ratio of the ordered-to-random mean-squared field amplitudes. We point out that variable afterglow light curves should be accompanied by a variable polarization. The radiation from the original ejecta, which includes the prompt gamma-ray emission and the emission from the reverse shock (the " optical flash" and " radio flare"), could potentially exhibit a high degree of polarization ( up to similar to 60%) induced by an ordered transverse magnetic field advected from the central source.