The cosmic microwave background (CMB) maps obtained by observations always possess domains, which have to be masked due to severe uncertainties with respect to the genuine CMB signal. Cosmological analyses ideally use full CMB maps in order to get, for example, the angular power spectrum. There are attempts to reconstruct the masked regions at least at low resolutions, that is, at large angular scales, before a further analysis follows. In this paper, the quality of the reconstruction is investigated for the ILC (7-yr) map as well as for 1000 CMB simulations of the Lambda cold dark matter concordance model. The latter allows an error estimation for the reconstruction algorithm, which reveals some drawbacks. The analysis points to errors of the order of a significant fraction of the mean temperature fluctuation of the CMB. The temperature two-point correlation function C(v) is evaluated for different reconstructed sky maps, which leads to the conclusion that it is safest to compute it on the cut-sky.