The fatigue strength of small specimens is observed to be significantly higher than that of large components. The gradient concept according to Eichlseder only accounts for the geometrical size effect, whereas the statistical and technologcal size effects are not quantitatively determined. Within the scope of the present investigation the statistical and the technological size effect as well as the impact of the type of loading were investigated and a model for their consideration was derived. By means of S,N-curves obtained with specimens made of the quenched and tempered steel 34CrNiMo6 the gradient concept could be validated for all test runs.They showed a significant impact of the type of loading that was applied for the determination of the curve that forms the basis of the gradient concept. The cyclic fatigue resistance under reversed bending, tension-compression and rotating bending descends according to the described order. This impact is considered with multiplicative coefficients that lead to a parallel translation of the curve. Accompanying measurements of the residual stresses in the surface layer allowed the determination of the sensitivity factor for residual stresses and therefore the consideration of residual stresses in the model. The technological size effect was investigated with S,N-curves that were determined with small specimens cut out of the rod diameter 80mm at appropriate position. These rods were also used in order to investigate the size effect with specimens of diameter 50mm. When comparing these results with those obtained with small specimens made of separately quenched and tempered material it is observed that the technological size effect is almost twice as much as the statistical size effect. It has not only got an impact on the fatigue strength, but also on the number of cycles where the S,N-curve makes a transition towards a horizontal line. The slope parameter, however, is not affected. The technological size effect on the fatigue strength is considered in the gradient model by the multiplicative scale factor FT that leads again to a parallel translation of the curve that was determined with small specimens. The statistical size effect is reduced with increasing relative stress gradient chi* and is almost not further visible starting from values chi*>5mm-1 at stress concentration factors Kt that are relevant to practical experience.That's why there has to be implemented a chi* dependent term containing the size effect scaling factor FSt in th gradient model.