This paper reports the results obtained in 71 fertilizer experiments with cotton conducted from 1935-36 up to 1941-42 in private farms located on the principal soil types of the State of São Paulo's cotton belt. In this work were compared 0, 40 and 80 kilograms of P2O5, per hectare, as well as three levels of nitrogen and of potash, and three forms of phosphates. Nitrogen and potash were used respectively as Chilean nitrate and potassium chloride; phosphorus, as superphosphate, Rhenaniaphosphate or bone meal. All the fertilizers were added, as usual, to the furrows and slightly mixed with the soil just before sowing. As a rule phosphorus increased considerably the yields, so that in spite of some poor results the average effect was satisfactory. However, the response to the higher rate of application was only slightly higher than that of the lower. Superphosphate was superior to both Rhenaniaphosphate and bone meal, which gave practically equal results. Nitrogen and potash, although giving good results in about one third of the experiments, did not increase the yield or even depressed it in most of them, and their higher levels generally gave poorer results than the lower. Some of the soils were apparently well supplied with nitrogen and potash, and the excessive spacing then used no doubt contributed to lessen the need for nitrogenous and potassic fertilizers. The application of nitrogen at planting time and the consequent loss through leaching before the cotton plants could take it up, may also have contributed to many small responses to that nutrient. These factors do not explain, however, why moderate doses of nitrogen and potash so often depressed the yields. Nitrogen, potash and even phosphorus (chiefly when used as Rhenaniaphosphate) reduced the stands in many experiments, and these reduction were the principal causes of the negative effects in the yields. As the reductions of the stands were atributed to the application of the fertilizers in contact with the seed, it was concluded that, for evaluating their effect they must be applied by other method more efficient than that used in the experiments.