Multiples are traditionally treated as undesired noise, but they are also real reflections from the subsurface as primaries. Smaller reflection angles and longer travel paths usually make them provide more structural information and more balanced illumination than primaries. Instead of multiple elimination in conventional seismic data processing, the migration of multiples has drawn great attention in recent years. The most commonly used method is performed by replacing the source wavelet with the observed data and using separated multiples as the receiver wavefield to apply traditional migration algorithms. However, crosstalk artifacts caused by crosscorrelation of unrelated events severely degrade the image quality of multiples. We have analyzed the cause of artifacts followed by a novel proposal of migrating the multiples by separating surface-related multiples into different orders. First, we combine surface-related multiple elimination and the focal transform to do the separation of multiples in order. Then, crosstalk can be well-eliminated by migrating different-order multiples separately and stacking the separated images together. Taking advantage of reverse time migration, imaging of multiples can be greatly improved. Theoretical analysis shows that crosstalk artifacts can be well-eliminated by our method. Numerical and field data examples determined that our method can provide a greater amount of correct information for subsurface structures.