We propose a power filter Cp for linear reconstruction of the CMB signal fr
om one-dimensional scans of observational maps. This Gp filter preserves th
e power spectrum of the CMB signal in contrast to the Wiener filter which d
iminishes the power spectrum of the reconstructed CMB signal. We demonstrat
e how peak statistics and a cluster analysis can be used to estimate the pr
obability of the presence of a CMB signal in observational records. The eff
iciency of the G, filter is demonstrated on a toy model of an observational
record consisting of a CMB signal and noise in the form of foreground poin
t sources.