Correlation migration for structural imaging using Gaussian beams is described for the inversion of passively recorded teleseismic waves. Gaussian beam migration is based on an overcomplete frame-based representation of the seismic wave field and uses localized slant-stack windows of the data. Paraxial Gaussian beams are then utilized for the backpropagation of the seismic waves into the medium. The method can provide stable imaging of seismic data in smoothly varying background media where caustics and triplicated arrivals exist. We develop Gaussian beam migration for structural imaging, which allows for incident teleseismic waves from beneath the structure, using directly scattered or surface-reflected phases. The method is applied to synthetic data computed for a collisional-zone model, and the inversions show that the Gaussian beam migration can image structure using passive teleseismic data. The method is next applied to synthetic data that have been convolved with an observed pulse from the 1993 Cascadia experiment. The autocorrelation is computed and the autocorrelated data are migrated by using Gausian beams for direct P-to-S scattered waves and surface-reflected waves. Although the migrations of the autocorrelation data with the source pulse included have more noise than the migrations of the impulsive source data, the subduction zone structure can still be clearly identified in the migrated results without removal of the source pulses.