<p>In flood frequency analysis, the annual maximum series are often fitted to the generalized extreme value (GEV) distributions. Given the uncertainty of hydrological processes and the finiteness of observed data, the results are usually expressed via confidence intervals based on quantile calculations. Previous studies have shown that errors in unilateral interval (quantile) estimates obtained by existing methods may sometimes lead to unreliable bilateral interval estimates. Therefore, in this paper, we present a novel approach based on pivotal quantities. The confidence interval estimations of the parameters and return levels are obtained by a ternary joint pivotal quantity group, specially formulated for the parameters in GEV distributions. The reliability of the proposed approach is evaluated through simulated experiments, and its performance is compared with other methods, including the test inversion bootstrap method, the profile likelihood method, and the standard bootstrap method. The pivotal quantity method showed great accuracy in estimating quantiles and confidence intervals of parameters. In the confidence interval estimations of return levels, compared to the other three methods, the pivotal quantity method is even more favored in the case of small sample sizes and high return periods, making it more practically applicable. The proposed approach is recommended for positively skewed GEV, Gumbel, and negatively skewed GEV distributions. The three specially formulated pivotal quantities can also be extended further to other three-parameter distributions commonly utilized in flood frequency analysis.</p>