We derive an optimal linear filter to suppress the noise from the COBE
Differential Microwave Radiometer (DMR) sky maps for a given power sp
ectrum. We then apply the filter to the first-year DMR data, after rem
oving pixels within 20-degrees of the Galactic plane from the data. We
are able to identify particular hot and cold spots in the filtered ma
ps at a level 2 to 3 times the noise level. We use the formalism of co
nstrained realizations of Gaussian random fields to assess the uncerta
inty in the filtered sky maps. In addition to improving the signal-to-
noise ratio of the map as a whole, these techniques allow us to recove
r some information about the cosmic microwave background anisotropy in
the missing Galactic plane region. From these maps we are able to det
ermine which hot and cold spots in the data are statistically signific
ant, and which may have been produced by noise. In addition, the filte
red maps can be used for comparison with other experiments on similar
angular scales.