Geotechnical parameters scatter in a wide range. On the one hand, this is due to the spatial variability of the subsoil, but also results of laboratory tests on reconstituted specimens of one sample scatter, as a completely homogeneous, reproducible specimen preparation is not feasible. For calculations according to the standards, characteristic shear parameters should be chosen as cautious estimate of the mean values. How this cautious estimate should be determined is not defined and therefore subjective. Often the results of shear tests are used as basis for the decision. In this paper, results of drained triaxial compression tests on a reconstituted, natural, widely graded soil are investigated. The specimens were prepared at same mean density but the results of the shear experiments scatter. The deviation of e.g. the peak strength is apparent. For the derivation of the Mohr-Coulomb parameter friction angle and cohesion according to the standards 3 or more stress levels have to be considered. The influence of the number of stress levels taken for the evaluation of the shear strength parameters is quantified. The evaluation of only three stress levels leads to a relatively large range of possible shear parameters. The two shear parameters friction angle and cohesion are statistically dependent - since they are two parameters of a linear regression. Therefore, they should be considered together. The scattering in the peak strength is probably caused by an inhomogeneous specimen construction. The influence of an inhomogeneous specimen preparation on the peak strength is investigated and proven in numerical simulations.