Computer simulations of the elongation cycle of bacterial protein biosynthesis demonstrate that the accuracy of protein biosynthesis cannot be explained by a mechanism which involves only an initial selection and a proofreading reaction. It is suggested that only a combination of initial selection, proofreading and a retardation of non-cognate flows at the level of the EF-Tu-catalyzed GTPase reaction and the peptidyl transfer can guarantee sufficient accuracy at reasonable costs. According to this view the ribosome functions as an allosteric enzyme which, in both its affinity and enzymatic activity, responds optimally only to the cognate substrate. Detailed calculations show, furthermore, that increasing the concentration of EF-G and EF-Ts above the level prevailing in vivo only slightly increases the rate of elongation. In contrast, increasing the concentration of EF-Tu over aminoacyl-tRNA (aa-tRNA) leads to a sharp decline in the rate of elongation. While varying the concentration of EF-G has no effect on the accuracy of protein synthesis, excess of EF-Tu over aminoacyl-tRNA leads to a large increase in accuracy. These results suggest a mechanism by which the accuracy of protein biosynthesis is preserved during amino acid starvation.