Reliable estimates of flood quantiles are essential for the design of dams, spillways, flood-control structures and reservoir operating rules, yet the parametric probability distributions conventionally used for flood frequency analysis can fit annual peak-flow data poorly when the sample is small, multimodal or heavily skewed. This study compares ten parametric distributions (two- and three-parameter log-normal, two-parameter gamma, Pearson and log-Pearson type 3, Gumbel/extreme value type I, logistic, log-logistic, generalized logistic and generalized log-logistic) with four nonparametric approaches (normal, log-normal and triangular kernel functions, and the Fourier series method) for the frequency analysis of annual peak flows at six hydrometric stations in Khouzestan province, Iran, with record lengths of 43 to 47 years. Parameters of the parametric distributions were estimated by both the method of moments and maximum likelihood, and the smoothing parameter of the kernel functions was obtained by the rule of thumb, the Sheather-Jones and the least-squares cross-validation methods. The goodness of fit of all fourteen methods was assessed using the mean relative deviation (MRD) and the mean square relative deviation (MSRD) between observed and computed flood values. The log-Pearson type 3 distribution gave the best fit among the parametric distributions for five of the six stations, and the generalized log-logistic distribution gave the best fit for the remaining station; among the three kernel functions, the log-normal kernel performed best. However, the Fourier series method outperformed all ten parametric distributions and all three kernel functions at every station, with substantially lower MRD and MSRD values. Because the Fourier series method requires no assumption about the underlying probability distribution and no choice among competing parameter-estimation procedures, it is proposed as a robust, easily applied alternative for flood frequency analysis in the study region and in other basins with similarly short or irregularly shaped flood records.