Grazing trials were established at two sites in the semi-arid savanna (Lowveld) of KwaZulu-Natal. The sites differed initially in rage composition. Llanwarne was dominated by Themeda triandra, Panicum maximum and P. coloratum and Dordrecht by Urochloa mosambicensis, Sporobolus nitens and S. iocladus. Three treatments at each site were stocked with Brahman-cross cattle to initially represent ‘light’ (0.17 LSU.ha-1) ‘intermediate’ (0.23 LSU.ha-1) and ‘heavy’ (0.30 LSU.ha¬-1) stocking. Cattle mass data collected over 116 three-week periods were used to develop a step-wise multiple linear regression model where summer mass gain (kg.ha-1) was significantly related (p < 0.01) to total seasonal rainfall (mm) (measured 1 July to 30 June) and stocking rate (LSU.ha-1). Winter mass loss (kg.ha-1) was related to residual herbage mass at the end of summer (kg.ha-1) and the length of winter (days). Although range condition did not significantly influence summer mass gain, winter mass loss was inversely related to residual herbage at the end of summer, which suggested that grass species in the Lowveld might differ in production potential rather than in quality.