Finding the optimum levels of nutrients in agricultural experiments is being used largely in recent years with the purpose to increase the yield associated with the maximum net income. The objective of the dissertations was to analyses the results obtained in fertilizer experiments and cane yield, sugar yield (t/ha), probable sugar yield (% of cane), adjusting to them a response, surface. Data make useful deriving from six mills in state of São Paulo, in which on incomplete factoring of FA0 type of seven points was designed. The treatment combinations were: 011; 111; 211; 101; 121; 110; 112; with two replicates. Two models were adjusted, say, quadratic polynomial regression and asymptotic Mitscherlich regression: Y_i = a₄₄ + a₁₄X_1i + a₂₄X_2i + a₃₄X_3i + a₁₁X²_1i + a₂₂X²_2i + a₃₃X²_3i + e_i Y_i = A [1- 10^{-c(x_i + b)}] + ε_i For the polynomial adjustment the individual and pooled analyses were performed when the following cases were considered: - Cane yield (t/ha) - Probable sugar yield (% of cane) - Sugar yield (t/ha). The main points treated were: a. Parameters estimates and interval estimates. b. Optimum levels of nutrients and their confidence intervals. c. Surface response cuts, fixing 2 levels in their average values, objecting an economical study. ln the Mitscherlich model adjustment the three cases mentioned above were also considered. The most important conclusions were: a. The confidence intervals for the polynomial model were larger than one could afford for practical purposes, and the quadratic terms were not significant in any case. Larger doses are required to reach better estimates. b. For the economical doses to be confiable it is necessary that the quadratic terms coefficients differ from zero statistically which was not verified in our case, turning the recommendation impossible. c. Incomplete factorial designs of FAO-type applied to fertilizers experiments in sugar cane showed no rational results interpretation when polynomial and asymptotic models were used.