The current clinical procedure for pure-tone audiometry was analysed for statistical measurement errors. Theoretically, the root-mean-square (RMS) error in a single threshold measurement is always greater than the standard deviation (SD) of measured intra-individual test-retest differences, divided by the square root of two. The RMS error includes an additional quantization component, caused by the finite step size between presented signal levels. In Monte-Carlo simulations with 2-dB and 5-dB steps the quantization error was negligible compared with other errors. Therefore, the single-test RMS error can be estimated with sufficient accuracy from the test-retest SD. The simulated single-test RMS error decreased from about 2.7 dB about 2.3 dB when the audiometric step size was reduced from 5 dB to 2 dB. Hearing thresholds appeared to be about 1.7 dB better with 2-dB steps than with 5-dB steps.