We illustrate the use of Bayesian strategy for estimating extreme hurricane winds that affect the United States based on separating the reliable 20^th century records from less precise 19^th century records. The analysis makes use of extreme value theory, so we assume that the maximum winds at landfall follow a generalized Pareto distribution. Wind information from 19^th century storms is used to obtain a prior on the two parameters of the Pareto distribution under the assumption that the prior is multivariate normal. The prior mean and variance are obtained using a bias-corrected bootstrap of the maximum likelihood estimates for the prior parameters. We use a Markov chain Monte Carlo algorithm to sample the posterior distribution. Posterior samples of the parameters are used to estimate a return period distribution at various levels of maximum winds for land falling storms. Our results show significant differences in these parameters for different climate regimes (ENSO, NAO). We conclude that the earlier records, though less reliable, allow for a more precise description of extreme hurricane winds.