The methodology of surface?wave retrieval from ambient seismic noise by crosscorrelation relies on the assumption that the noise field is equipartitioned. Deviations from equipartitioning degrade the accuracy of the retrieved surface?wave Green's function. A point?spread function, derived from the same ambient noise field, quantifies the smearing in space and time of the virtual source of the Green's function. By multidimensionally deconvolving the retrieved Green's function by the point?spread function, the virtual source becomes better focussed in space and time and hence the accuracy of the retrieved surface?wave Green's function may improve significantly. We illustrate this at the hand of a numerical example and discuss the advantages and limitations of this new methodology.