Problems arise with coastal flood modelling because of the short time scale of gauged records and the lack of observed data to constrain the model. Records going back, say 50 years, will not necessarily capture any extreme coastal flood events and even where such an event has been recorded, there are few instances where a comprehensive data set exists to determine the hydraulic boundary conditions for calibration and the flood extent for validation of the model. Whilst the use of historical data, predating gauged records, has been propounded in fluvial hydrology since the NERC 'Flood Studies Report' of 1975, there have been no previous coastal flood modelling studies in the UK which use such historical data to constrain the model. This thesis proposes to evaluate whether historical data can be used in the context of coastal flood modelling and whether the use of multiple validation data sets can offset the problems of imprecision in the data. A chronology of severe storm and coastal inundation events in the Bristol Channel and Severn Estuary is presented. The archival research revealed many more historical sea-floods than had been anticipated. Whilst the 1607 inundation has widely been held as the most severe to have ever occurred in the Severn Estuary, seven other events of similar, if not greater, magnitude have been documented in the last 500 years. The apparent increase in frequency of coastal inundation events can be explained by the increased reporting since the mid-1800s. There is no evidence of increasing extreme tidal elevations. Selected historic floods were researched in depth and those with sufficient information on tidal elevations and flood extent were simulated using the LISFLOOD-FP inundation model. The knowledge gained from using multiple data sets outweighs the inherent uncertainties of individual cases and has enabled a reality-check on the more speculative estimates of depth and extent of past inundations. This thesis demonstrates the value, both of the research into the historical record for data on past flood events, and in the construction of a model to corroborate that data. This approach would be applicable in any project to investigate the historical record for the purpose of coastal risk assessment.