Transmitter power control has proven to be a powerful technique to control interference and improve the performance of wireless networks. Efficient power control normally re quires fast and reliable quality measurements. In cellular systems where propagation and interference conditions vary rapidly, such realtime measurements cannot be obtained. In stead, diversity techniques can be useful. Frequency hopping combined with error control coding is an example how this diversity can be implemented. These systems effectively ''average'' the interference. How power control should be applied in these systems is, however, not obvious. In this paper, we show that symbol error performance in the uplink of a coded fre quency hopping cellular system is, at least in principle, minimized by maintaining the received signal power at a constant level, provided that accurate instantaneous measure ments are available. When such instantaneous measurements are not available, but only time averages this is no longer the optimum strategy. Therefore, also some signallevel based power control schemes are analyzed with respect to their interaction with the error control coding schemes used. Results show that power control schemes designed for static channel allocation (without interference randomization) may give rise to severe fairness problems in coded frequency hopping systems. It is also found that changing the code parameters does not significantly affect the choice of power control policy.