The next generation of wireless networks will require efficient use of the underlying time-varying channel. Cellular architectures efficiently utilize the limited spectrum by reusing the same channel in spatially separated cells. However, frequency reuse introduces cochannel interference, which determines the data rate that can be supported by each channel for a given BER. The efficiency within a cell is quantified by its spectral efficiency: the data rate per user per unit of bandwidth. In this paper, the maximum spectral efficiency for cellular systems with channel estimation and transmitter feedback is derived from Shannon theory. We first review capacity resultsfor single-user time-varying channels, which show that eficiency is inax-imized through a combination of power control, adaptive coding, and variahle data rates. We then extend this result to the multiuser capacity region for both broadcast and multiaccess channels. 7Ris yieMs an analytical framework to compare the efficiency of channel sharing within a single cell using TDMA, FDMA, and CDMA. We conclude that CDMA with interference cancellation is optimal, but TDMA and FDMA have superior performance if interference cancellation is not exploited. Moreover, i f the received power of all users is kept constant, then CDMA with interference cancellation does no better than FDMA and TDMA.