Responses of yellow serradella, slender serradella and subterranean clover to applications of superphosphate and Queensland apatite rock phosphate were compared in a field experiment in Western Australia. The rock phosphate was applied as a fine powder to a very sandy soil with a low buffering capacity for phosphate and for pH. At low levels of application, rock phosphate was about a tenth as effective as superphosphate - that is, ten times as much was required to give the same yield. With increasing levels of application, the relative effectiveness of rock phosphate declined to well below one per cent and hence more than 100 times as much was required to give the same yield. The decreasing relative effectiveness was incorporated into response equations. These were used to demonstrate that the decreasing relative effectiveness can be statistically tested and shown to be highly significant. A simulation study using the model of Kirk and Nye showed that a decrease in relative effectiveness with increasing level of application is especially likely when the particles of rock phosphate are very small, when the levels of application are high, and when the soil's buffering capacity for phosphate and for pH is low.