Multiuser detection (MUD) of signals occupying the same time/frequency slot has been shown to provide significant increases in the capacity of a single cell or coverage area. In a cellular system, however, increasing the number of users per cell also produces increased interference from other cells. Compensation by increasing the co-channel cell spacing partially offsets the capacity gain due to MUD, and the net effect on system capacity then rests on the detection method, such as maximum likelihood (ML) or minimum mean-squared error (MMSE). This paper is the first to analyze the net capacity gains due to ML-MUD in time division multiple access (TDMA), and the first to compare ML-MUD with MMSE-MUD on this basis. We demonstrate that ML-MUD with M cochannel users per cell and four base station antennas gives a net system capacity that is M times greater than that of single-user TDMA system with two antennas. In contrast, we show that MMSE-MUD gives only a modest net capacity increase, and that the system capacity may actually decrease with more than one or two users per cell, making it poorer than the original single-user system.