We report the results of a statistical analysis performed with the four foreground-cleaned Planck maps by means of a suitably defined {\it local-variance} estimator. Our analysis shows a clear dipolar structure in Planck's {\it variance map} pointing in the direction $(l,b) \,\simeq\, (220^{\circ},-32^{\circ})$, thus consistent with the North-South asymmetry phenomenon. Surprisingly, and contrary to previous findings, removing the CMB quadrupole and octopole makes the asymmetry stronger. Our results show a maximal statistical significance, of $98.1\%$ CL, in the scales ranging from $\ell=4$ to $\ell=500$. Additionally, through exhaustive analyses of the four foreground-cleaned and individual frequency Planck maps, we find unlikely that residual foregrounds could be causing this dipole variance asymmetry. Moreover, we find that the dipole gets lower amplitudes for larger masks, evidencing that most of the contribution to the variance dipole comes from a region near the galactic plane. Finally, our results are robust against different foreground cleaning procedures, different Planck masks, pixelization parameters, and the addition of inhomogeneous real noise.

19 pages, 13 figures. Re-submitted to match published version