Simulation was used to evaluate the properties of (1/5) (5³) designs, obtained by the superposition of three orthogonal Latin squares. The following basic quadratic equation was used: Yijk = 3500 + 180 x li+ 250 x lj + 120 x lk - 42 x 2i - 55 x2j - - 28 x2k - 25 x lilj - 18 x lilk- 12 x ljlk (A) The values of the linear and second order polynomials for the quadratic model and the correspondent polynomials for the square root model were, respectively: Quadratic: g1=-3+X, g2 =7- 6X+X² Square root: x?= a???+(X)½ e x2=a2+ g2 (X + X,)½ where a?=1.67646; a2=2.41157; g2=-3.22798, obtained with the restrictions for orthogonality: Sx?= Sx2= Sx?Sx2,=0, as described in the "Designs (1/5) (5x5x5)", given by the authors. The coefficients given in (A) represent the mean value, the linear, quadratic, and interactions effects of three factors npk from a surface response expressed in kg of corn per hectare; their values were chosen in such a way that main effects were significantly different from zero, the quadratic coefficients and interactions allowing a maximum point of the function to be allocated between the fourth and fifth levels, so inside the range of the dosages used. This is the situation that should be looked for when choicing the levels of the dosages in planning field fertilizer experiment programmes: we must avoid dosages that give origin to plateau response because they are far apart from the area of economical decision. A histogram of normal shape for the errors, using a coefficient of variation of 8.5%, was built, and the errors were distributed at random for the 25 treatments in the equation (A), creating 60 sets of data. The quadratic and square root polynomials were fitted to each set of 25 treatments and also to the mean of the treatments obtained from groups of n experiments, n equal to 2, 3, 4, 5, 6, 10, 12, 15, 20 or 60. A good convergence of the estimators of the parameters was obtained for the two models. The coefficients of determination were practically of the same value. The percentage of maximum, obtained through the canonical form for both models for the different groups, is: MODEL GROUP MODEL 1 2 3 4 5 6 10 12 15 20 30 60 Quadratic 35 70 70 67 92 90 100 100 100 100 100 100 Square root 13 27 35 47 50 50 67 80 75 100 100 100 The coefficients of the variance of the treatments on the main diagonal were: MODEL TREATMENT 1 2 3 4 5 6 10 12 15 20 30 60 Quadratic 33 70 65 67 92 90 100 100 100 100 100 100 Square root 15 27 35 47 50 50 67 80 75 100 100 100 The results indicated that the grouping of 10 experiments of the (1/5) (5x5x5) design was sufficient to obtain 100 percent maximum, when quadratic model was used, whereas 20 experiments are necessary when the square root model was utilized. Another basic equation (C), related to (A) but of square root nature, was used for a similar simulation. The results obtained for bP and the percentage of maximum points were analogous. Independently of the basic equation (A) or (C), with a coefficient of variation of 8.5%, the grouping of 10 experiments was the first one that assured the obtaining of 100 percent maximum, when quadratic response surface was fitted, whereas 20 experiments were necessary when the square root model was used.