A systematic, or continuous, at-site record of annual maximum flood peaks can be readily combined with other information to predict flood quantiles and their standard errors using a Bayesian Markov Chain Monte Carlo inference approach. Sources of additional information include records of historic flood peaks, results of regional flood frequency analyses, output from rainfall-runoff models and prediction of flood peak magnitudes derived from analogue basins having similar flood hydrology. The approach is illustrated by a case study using the Waimakariri River at Old Highway Bridge site, which has a long systematic and historic flood record and regional, model and analogue information. Various periods of the systematic record combined with some and all of the additional information show that as more information is included in the analyses, the standard errors of predicted values of both 100 and 300-year return period flood peaks reduces markedly. While further work on sources of other information is desirable, including, in particular, paleohydrological investigations, the Bayesian inference approach should improve prediction of flood quantiles and their uncertainty for at-site, regional and national studies.