Abstract. We describe polarization observations of the CMBR with the Cosmic Background Imager, a 13 element interferometer which operates in the 26-36 GHz band from Llano de Chajnantour in northern Chile. The array consists of 90-cm Cassegrain antennas mounted on a steerable platform which can be rotated about the optical axis to facilitate polarization observations. The CBI employs single mode circularly polarized receivers which sample multipoles from £~400 to ^4250. The instrumental polarization of the CBI was calibrated with 3C279, a bright polarized point source which was monitored with the VLA.

We present new measurements of the power spectra of the E-mode of CMB polarization, the temperature T, the cross-correlation of E and T, and upper limits on the B-mode from 2.5 years of dedicated Cosmic Background Imager (CBI) observations. Both raw maps and optimal signal images in the uv-plane and real space show strong detections of the E-mode (11.7 sigma for the EE power spectrum overall) and no detection of the B-mode. The power spectra are used to constrain parameters of the flat tilted adiabatic Lambda-CDM models: those determined from EE and TE bandpowers agree with those from TT, a powerful consistency check. There is little tolerance for shifting polarization peaks from the TT-forecast locations, as measured by the angular sound crossing scale theta = 100 ell_s = 1.03 +/- 0.02 from EE and TE cf. 1.044 +/- 0.005 with the TT data included. The scope for extra out-of-phase peaks from subdominant isocurvature modes is also curtailed. The EE and TE measurements of CBI, DASI and BOOMERANG are mutually consistent, and, taken together rather than singly, give enhanced leverage for these tests.Comment: 15 pages, 9 figures, submitted to ApJ -- Accepted version. The fine-bin spectrum, covariance matrix, and window functions are now available on the web (suitable for use in COSMOMC) at: http://www.astro.caltech.edu/~tjp/CBI/data2006/index.html The pipeline in the previous version inadvertently omitted one antenna, so the new spectrum contains ~15% more data. We emphasize that previous results were in no way biased, and that the (small) changes to the spectrum solely reflect the inclusion of the additional data. Numbers and figures in the paper have been updated correspondingly. All maps now have color bar