Abstract The chemical composition of the highest-energy cosmic rays, namely the atomic number Z of rays with energies E ≳ 40 EeV, remains to date largely unknown. Some information on the composition can be inferred from the deflections that charged ultra-high-energy cosmic rays experience while they traverse intervening magnetic fields. Indeed, such deflections distort and suppress the original anisotropy in the cosmic ray arrival directions; thus, given a source model, a measure of the anisotropy is also a measurement of the deflections, which in turn informs us on the chemical composition. In this work, we show that, by quantifying ultra-high-energy cosmic ray anisotropies through the angular cross-correlation between cosmic rays and galaxies, we would be able to exclude iron fractions f Fe ≥ (10%) assuming a fiducial hydrogen map at 2 σ level, and even smaller fractions in the reverse case of hydrogen on an iron map, going well below f H ≈ 10% when we mask the Galactic Centre up to latitudes of 40°. This is an improvement of a factor of a few compared to our previous method, and is mostly ascribable to a new test statistics which is sensitive to each harmonic multipole individually. Our method can be applied to real data as an independent test of the recent claim that current cosmic-ray data can not be reproduced by any existing model of the Galactic magnetic field, as well as an additional handle to compare any realistic, competing, data-driven composition models.