Seismic migration is the process by which wavefields recorded on or near the surface are mapped back into the subsurface to form an image of the subsurface structure. Common to all current migration methods is that they rely on a model of the subsurface for computing Green's functions. These computed Green's functions are approximations of the true Green's function in the subsurface. The more accurate the method used for computing the Green's functions, and to some extent also the choice of imaging condition, the better image we will get. Most state-of-the-art migration methods will use Green's functions that include multiple arrivals and some finite-frequency effects, allowing them to image fairly complex geologic structures. However, the computational model in most cases uses only scalar velocity that is smooth on a wavelength scale, thus ignoring scattering effects from small-scale structure, anisotropy, and other intrinsic rock effects. Further, whatever the model is, it has to be derived from the data itself or some other form of remote measurement of velocity in the subsurface. At the very best, this will be a nonunique solution to an inverse problem based on the data. Commonly, much more crude approximations are used; for example, only trying to match the predicted first arrivals to the observed first arrivals.