On 2 October 2004, a significant noneruptive tremor episode occurred during the buildup to the 2004–2008 eruption of Mount St. Helens (Washington). This episode was remarkable both because no explosion followed, and because seismicity abruptly stopped following the episode. This sequence motivated us to consider a model for volcanic tremor that does not involve energetic gas release from magma but does involve movement of conduit magma through extension on its way toward the surface. We found that the tremor signal was composed entirely of Love and Rayleigh waves and that its spectral bandwidth increased and decreased with signal amplitude, with broader bandwidth signals containing both higher and lower frequencies. Our modeling results demonstrate that the forces giving rise to this tremor were largely normal to conduit walls, generating hybrid head waves along conduit walls that are coupled to internally reflected waves. Together these form a crucial part of conduit resonance, giving tremor wavefields that are largely a function of waveguide geometry and velocity. We find that the mechanism of tremor generation fundamentally masks the nature of the seismogenic source giving rise to resonance. Thus multiple models can be invoked to explain volcanic tremor, requiring that information from other sources (such as visual observations, geodesy, geology, and gas geochemistry) be used to constrain source models. With concurrent GPS and field data supporting rapid rise of magma, we infer that tremor resulted from drag of nearly solid magma along rough conduit walls as magma was forced toward the surface.