The effect of using the plotting position formulae (PPF) of(1) Landwehr, (2) (i - 0.35)/n, (3) Cunnane, (4) Weibull for computation of probability-weighted moments (PWMs) on the parameters of the probability distributions (PDs) of three-parameter log-normal, log-Pearson-3, general extreme value, and Wakeby by the method of PWMs has been experimentally investigated. The method of maximum-likelihood has also been applied on the three three-parameter PDs. Altogether 19 models were formed. The prediction of the three right-tail quantiles of T = 100, T = 1000, and T = 10 000 year return-period peaks by any PD has been determined by comparing the distributions of the 1000 relative errors of each one of these three extreme values computed out of 1000 short synthetic samples from the same quantiles of the parent distributions. Inspection of box plots of such computed relative errors led to the conclusion that the general extreme value and log-Pearson-3 PDs applied by the method of PWMs using the Landwehr PPF are the better models of the 19 considered. Three-parameter log-normal distribution either by the method of maximum-likelihood or by the method of PWMs using the PPF of either (i - 0.35)/n or Landwehr turned out to be almost as good as those two. Convincing evidence for superiority of the Wakeby PD over these three three-parameter PDs has not been found.