The knowledge of moisture-content changes in shallow soil layers has important environmental implications, and ground-penetrating radar (GPR) used in surface-to-surface configuration has been used increasingly to quickly image soil moisture content over large areas. The technique is based on measuring direct GPR wave velocity in the ground. However, in the presence of shallow and thin low-velocity soil layers, dispersive guided GPR waves are generated and the direct ground wave is not identifiable as a simple arrival. Under such conditions, the dispersion relation of guided waves can be estimated from field data and then inverted to obtain the properties of the guiding layers. This approach is applied to a mountain slope with a 1-m soil cover where repeated measurements over time, inverted by conceptualizing the soil as a single guiding layer, lead to estimates of velocity andthickness varying over time. Varying soil thickness clearly is not a plausible physical process. To remove this problem, we develop a multilayer GPR waveguide model. We first assess, using a Monte Carlo sensitivity analysis, the model error arising from using a single-layer forward model to invert data generated by a multilayer waveguide. The single-layer model always underestimates the total soil thickness because the inversion is sensitive mainly to the layer with the lowest velocity (the wettest layer). We then use a multilayer forward model to invert the actual field data. By constraining the total soil thickness, we still manage to invert accurately only for velocity and thickness of the wettest layer, leaving uncertainty about the position of such a layer in the layer sequence. We conclude that these inversion equivalence problems cannot be neglected when guided GPR data are used to estimate time-lapse moisture content in shallow soils.