Stolt migration is often performed with many different velocities, creating a cube (or ensemble) of migrated constant‐velocity sections. With this ensemble, an interpreter can quickly, even interactively, carve out a single, variable‐velocity time migration. Unfortunately, the variable‐velocity section obtained in this way cannot image steep dips correctly. Phase‐shift (i.e., Gazdag) migration accurately images reflections from steep interfaces where velocity varies with depth. However, it is computationally slow compared to Stolt’s f-k migration, its constant‐velocity counterpart. Similarities between the Gazdag and Stolt methods allow a mapping between the two to be derived. This new mapping, which replaces the constant‐velocity Stolt mapping, defines a new Stolt‐like migration. Stolt‐like migration can then be performed to create an ensemble of migrated data, from which a Gazdag‐equivalent migration can be carved. All dips can be imaged properly, even where velocity varies significantly with depth, while the advantages of working with the ensemble of migrated sections are preserved.