Abstract: Historical flood-concerned mixed-distribution method (HFCMM) not only extends sample length of flood runoff data but improves the traditional flood frequency analysis lacking physical ground. This study applied this method to Annual Maximum Flood(AMF) series from the Wujiang basin and developed a mixed-distribution model of annual maximum flood runoff, using historical flood-concerned L-moment for parameter estimation. Comparison of the modeling results for different Probability Distribution Function(PDF) tail types shows that HFCMM increases the recurrence intervals of high flood flows and improves their flood frequency analysis by adopting heavy-tailed or three-parameter mix-tailed PDFs but not two-parameter mix-tailed PDFs. But with thin-tailed PDFs fitting, the method reduces the recurrence intervals of low flood flows, and thus this fitting is not recommended. For flood frequency analysis of the Wujiang basin by HFCMM, the distributions of GLO and GEV fit better than the traditional Pierson III.