For the estimation of the unknown parameters θ in the regression model   it will be given a nonsequential method of experimental design in this paper. In the case that f(x, θ) depends on θ in a nonlinear way the information matrix   of the continuous normed designed ξ depends on the the unknown parameters θ. CHERNOFF considered local optimal designs in this case. Using the criterion of optimally   for a measure Q over a σ-algebra of Θ we succced now in constructing designs that have some desirable properties globally for all θ ∈ Θ respectively Q. An iterative computing procedure and some properties of S-optimal designs are given. To simplify the computation ε-S-optimal designs are considered, i.e. designs that are approximately S-optimal in some sense. For this designs it will be shown that there are ε-S-optimal designs with a finite spectrum.