Yeast aminoacyl-tRNA synthetases act in a multi-step process when recognizing their cognate amino acids; this identification event includes "physical" binding and "chemical" proof-reading steps. However, the various enzymes use these single steps at different degrees, and their specificities with regard to the 20 naturally occurring amino acids deviate considerably from each other. The characteristic discrimination factors D were determined for seven synthetases in vitro: the highest specificity with D values between 28,000 and > 500,000 were observed with tyrosyl-tRNA synthetase, the lowest values between 130 and 1700 for lysyl-tRNA synthetase. The tested class I enzymes are more specific than the investigated class II enzymes, and it may be put into discussion whether this observation can be generalized. Error rates in amino acid recognition differ not only between the individual aminoacyl-tRNA synthetases but also considerably for different amino acids sorted by the same enzyme. Strikingly, all investigated enzymes exhibit a poor specificity in discrimination of cysteine and tryptophan from their cognate substrates, and these cases may be regarded as "specificity holes". In view of the observed specificities a protein consisting of 700 amino acids would contain maximally up to five "incorrect" residues, if the in vitro error rates are also valid under in vivo conditions. Therefore the terminus "quasi-species", an expression which was originally created for nucleic acids, is justified. The "quasi-species" nature of proteins may become important when genes are translated in different organisms with different accuracies of the translation apparatus. In such cases different "quasi-species" will be obtained. Using our data in mathematical models which predict the stability of protein synthesizing systems, we find that they are consistent with a stable yeast organism which is not prone to die by an "error catastrophe". However, this appears only if average values from our experiments are used for calculations. If a single compound, e.g. the arginine analog canavanine, is discriminated very poorly from the cognate substrate, or if the "specificity holes" get larger, an "error catastrophe" must be envisaged.