With the rapid development of the petroleum industry, oil and gas exploration gradually changes from conventional shallow and middle-depth reservoirs to deep and ultradeep reservoirs. In western China, especially in the Tarim Basin, the deep Ordovician and ultradeep Cambrian carbonate is the most important hydrocarbon reservoir. But because of complex near-surface conditions and complicated subsurface structures, high-precision seismic imaging for such deep and ultradeep reservoirs is still challenging with the state-of-art migration methods. One of the critical factors is that the reflections from deep and ultradeep strata cannot be coherently stacked in the migration because of accumulative traveltime errors caused by an inaccurate velocity model. To mitigate this issue, we present a coherent-stack-based least-squares migration (LSM) approach to improve the imaging quality for deep and ultradeep structures. Unlike traditional LSM that uses the stacked gradient during iterations, the proposed method updates the reflectivity model in the subsurface half-opening angle domain and then applies a coherent stacking to implement constructive summation for angle-domain common-image gathers. The new LSM scheme enables us to reduce the artifacts caused by an inaccurate velocity model and produces high-quality images for deep and ultradeep strata. Two models with typical steep-dipping faults, overthrust folds, and fault-karst carbonate reservoirs are designed to test the feasibility of the proposed method, and a field data set from a land survey is used to demonstrate its adaptability for low signal-to-noise ratio data.