We investigate observationally and theoretically the response of polarimetric backscattering at 24-cm wavelength to the thickness of Arctic sea ice in leads and first-year ice features. We employ backscattering data acquired by the Jet Propulsion Laboratory airborne synthetic aperture radar (SAR) during March 1988 in the Beaufort Sea, together with nearly simultaneous passive microwave imagery acquired by the U.S. Navy Ka band radiometric mapping system. We find that 24-cm copolar ratios and copolar phases vary strongly with apparent ice thickness. We observe copolar phase shifts between −10° and −50° (relative to multiyear ice phases) for new ice features in the imagery, as well as positive copolar phases in a first-year ice feature. Copolar ratios also vary with apparent thickness, from values larger than those expected theoretically for seawater to values slightly lower than those expected for thick ice. We derive a signature model based on scattering from a rough air/sea ice interface with realistic vertical profiles of brine volume and relative permittivity beneath. Model predictions for copolar ratios and phases show ice thickness-dependent variations consistent with those observed. We present simulation results showing that plausible ice thickness variations between pixels in a multilook average diminish, but do not eliminate, the signature response to thickness. This suggests that direct thickness estimation of sea ice in leads may be possible using polarimetric SAR at wavelengths of 24 cm or longer.