Guided waves, such as Lamb waves, are attractive tools for monitoring large civil infrastructures due to their sensitivity to damage. Yet, interpreting guided wave data and identifying effects resulting from damage is often complicated by the multimodal and dispersive characteristics of guided waves and multipath interference from the medium's boundaries. In this paper, we present a method to decompose guided waves into a collection of multipath arrivals by combining sparse wavenumber analysis, a methodology for accurately recovering multimodal and dispersive properties, with additional 1 minimization techniques. Its application to experimental Lamb wave data shows that the estimates all correspond to expected paths.