The reproduction of colors for the CIE standard observer is definitely determined if display devices with three color channels like a CRT or a TFT display are used. In contrast, multiprimary displays provide several possibilities to reproduce colors for the CIE standard observer. Due to the higher number of variables the degrees of freedom can be used to optimize the color difference between color stimuli of the reproduction and the original for several different observers. The achievement of small reproduction errors for several observers is one of the major advantages, a multiprimary display offers in addition to an essential the enlargement of the color gamut.In this paper a stochastic algorithm is described to optimize the control of a 6-channel multiprimary display. The aim is to reproduce multispectral input data for any human observer at smallest possible color differences. The origin of the algorithm is the additive mixture of 6 narrow band spectral primaries, which are described by measured spectral power distributions of a laboratory model. The quality of the color reproduction is defined as the maximum color difference between the reference and reproduction for 24 different human observers characterized by their spectral color matching functions. The maximum color difference of the observers calculated in CIE ΔE₉₄ units is minimized as a function of the 6 control vectors of the primaries. In contrast to previous publications a pseudo gradient procedure is used.For each specific input spectrum the final control vector of the primaries is achieved by a series of stochastic variations of the control vector. During the modification of the control vector the increment of variation is adapted. This paper demonstrates the effectiveness of the algorithm and shows that the maximum color difference for a the set of 354 representative color stimuli is about ΔE_94max ≈ 1 considering e.g. D₅₀, D₅₅, D₆₅, or D₇₅ as reference white, and the mean color difference is about ΔE₉₄ ≈ 0.3 for the mentioned illuminants.