||In recent years, dam safety has drawn increasing attention from the public. This is because floods resulting from breaching of dams can lead to devastating disasters with tremendous loss of life and property, especially in densely populated areas. Past disasters showed that the consequences of dam failures are directly related to the evacuation time available should failures occur. It is therefore very important to study dam failures and their corresponding breaching characteristics, and if possible, to model the dam breaching process. In order to improve the safety of dams, professionals experienced in dam engineering have made continuing efforts on design, construction, operation, and maintenance of dams. However, many existing dams still pose increasing hazards to the downstream areas due to structural deterioration, inadequate design, faulty construction, and poor operation and maintenance. These dams are referred to as distressed dams. Obviously, proper diagnosis of dam distresses and their corresponding causes is essential to ensure the dam safety. Based on such understanding, appropriate remedial measures may be suggested for improving the safety of such distressed dams. In this thesis, performance data of 1609 cases of failed dams are collected by HKUST and 1182 cases of distressed dams are collected by the China Institute of Water Resources and Hydropower Research (IWHR) for study in a cooperative research with IWHR. Details of the characteristics of the dams as well as the failure or distress information are collected. These data are compiled into two databases: Database A for failed dams and Database B for distressed dams. Based on Database A, a statistical analysis method is used to study dam failures. Further, breaching characteristics of embankment dams are investigated according to the type of fill material. To quantitatively describe the dam breaching process, robust empirical formulas with physical meaning are developed for predicting dam breaching parameters through a multivariable nonlinear regression model. The balance between the prediction accuracy and simplicity has been fully considered in the development of the empirical formulas. The relative ability of each control variable to be a predictor of associated dam breaching parameters is compared. In addition, a simplified prediction method is proposed for quickly determining the outflow hydrograph from the breach. Two case studies on the failures of Banqiao Dam and Teton Dam are presented to show the use of these models to predict the breaching parameters and the outflow hydrographs from the breaches. Based on Database B, the characteristics of embankment dam distresses are studied. The technique of Bayesian networks is used to develop a robust probability-based tool for the diagnosis of dam distresses at the global level based on past performance records. Bayesian networks are well suited for studying a complex dam system, which can tackle not only the multiplicity of dam distresses and causes but also the various interrelations within them. The common patterns and causes of embankment dam distresses are identified. A sensitivity analysis is also conducted to identify the most important factors contributing to the distresses of the embankment dams. Based on prior common characteristics extracted from the global-level data, the Bayesian networks are further used to diagnose a specific distressed dam at the local level by combining global-level performance records and local-level performance records and/or analysis results. A case study on Chenbihe Dam is presented to illustrate the probability-based diagnosis methodology.