The sensitivity of interferometers with linear polarizers to the cosmic microwave background (CMB) E and B modes is variant under the rotation of the polarizer frame, while interferometers with circular polarizers are equally sensitive to E and B modes. We show analytically and numerically that the diagonal elements of window functions for CMB E/B power spectra are maximized in interferometric measurements of linear polarization, when the polarizer frame is in certain rotation from the associated baseline. We also present a simulated observation to show that the 1σ errors on E/B-mode power spectrum estimation are variant under the polarizer frame rotation in the case of linear polarizers, while they are invariant in the case of circular polarizers. A simulation of a configuration similar to the Degree Angular Scale Interferometer (DASI) shows that the minimum 1σ error on the B mode in interferometer measurements with linear polarizers is 26 per cent of that in interferometric measurements with circular polarizers. The simulation also shows that the E/B mixing in interferometer measurements with linear polarizers can be as low as 23 per cent of that in interferometric measurements with circular polarizers. It is not always possible to physically align the polarizer frame with all the associated baselines in the case of an interferometer array (N > 2). There exist certain linear combinations of visibilities, which are equivalent to visibilities of the optimal polarizer frame rotation. We present the linear combinations that enable B-mode optimization for an interferometer array (N > 2).