We present new full-sky temperature maps in five frequency bands from 23 to 94 GHz , based on data from the first three years of the \map sky survey .
The new maps are consistent with the first-year maps and are more sensitive .
The three-year maps incorporate several improvements in data processing made possible by the additional years of data and by a more complete analysis of the polarization signal .
These include several new consistency tests as well as refinements in the gain calibration and beam response models ( 90 ) .

We employ two forms of multi-frequency analysis to separate astrophysical foreground signals from the CMB , each of which improves on our first-year analyses .
First , we form an improved “ Internal Linear Combination ” ( ILC ) map , based solely on \map data , by adding a bias correction step and by quantifying residual uncertainties in the resulting map .
Second , we fit and subtract new spatial templates that trace Galactic emission ; in particular , we now use low-frequency \map data to trace synchrotron emission instead of the 408 MHz sky survey .
The \map point source catalog is updated to include 115 new sources whose detection is made possible by the improved sky map sensitivity .

We derive the angular power spectrum of the temperature anisotropy using a hybrid approach that combines a maximum likelihood estimate at low l ( large angular scales ) with a quadratic cross-power estimate for l > 30 .
The resulting multi-frequency spectra are analyzed for residual point source contamination .
At 94 GHz the unmasked sources contribute 128 \pm 27 \mu K ^ { 2 } to l ( l + 1 ) C _ { l } / 2 \pi at l = 1000 .
After subtracting this contribution , our best estimate of the CMB power spectrum is derived by averaging cross-power spectra from 153 statistically independent channel pairs .
The combined spectrum is cosmic variance limited to l = 400 , and the signal-to-noise ratio per l -mode exceeds unity up to l = 850 .
For bins of width \Delta l / l = 3 % , the signal-to-noise ratio exceeds unity up to l = 1000 .
The first two acoustic peaks are seen at l = 220.8 \pm 0.7 and l = 530.9 \pm 3.8 , respectively , while the first two troughs are seen at l = 412.4 \pm 1.9 and l = 675.2 \pm 11.1 , respectively .
The rise to the third peak is unambiguous ; when the \map data are combined with higher resolution CMB measurements , the existence of a third acoustic peak is well established .

Spergel et al .
( 149 ) use the three-year temperature and polarization data to constrain cosmological model parameters .
A simple six parameter \Lambda CDM model continues to fit CMB data and other measures of large scale structure remarkably well .
The new polarization data ( 127 ) produce a better measurement of the optical depth to re-ionization , \tau = 0.089 \pm 0.03 .
This new and tighter constraint on \tau helps break a degeneracy with the scalar spectral index which is now found to be n _ { s } = 0.958 \pm 0.016 .
If additional cosmological data sets are included in the analysis , the spectral index is found to be n _ { s } = 0.947 \pm 0.015 .