The problem of obtaining estimates for the cosmic microwave background
(CMB) anisotropy spectrum when applying the Galaxy-removal procedure
is discussed. A method of two-step spatial-frequency filtering that re
tains the orthogonality of spherical harmonics is proposed. This metho
d is used to analyze the two-year COBE data. After the analysis of the
difference maps containing only noise, the l = 7, 9, 13, 23, and 25 m
ultipoles were excluded from the data under consideration because the
noise power in these multipoles in the difference maps turned out to b
e statistically anomalous. This procedure is shown to introduce no sig
nificant systematic errors for the model noise-spectrum employed. Once
the multipoles that are anomalous in terms of noise have been removed
, the analysis yielded a power-law index of the CMB anisotropy power s
pectrum of n = 1.84 +/- 0.29 and a quadrupole mean of Q(2) = 15.22 +/-
3.0. It has been established that the COBE data are not described by
the Harrison-Zel'dovich spectrum at 99% confidence. The need to increa
se the instrument sensitivity in order to more reliably separate the c
osmological component from the microwave background is demonstrated.