Numerous flood disasters occur every year worldwide, mostly resulting from failure of dikes, some with catastrophic outcomes. Examples are the dike failures along the Yangtze River in China in 1931 due to heavy rainstorms and those in the Netherlands in 1953 due to a storm surge at the North Sea. Both floods induced many deaths and enormous economic losses. The consequences of a flood induced by a dike burst can vary strongly and depend mainly on the speed and rate of inundation of the polder. This inundation speed and rate depend strongly on the flow rate through the breach, which in its turn depends on the development of the breach in time. Comprehension of the breach development process, and the mathematical translation of this process into a model, is of great importance to the design method of dikes based on a riskapproach. It is also important for the development of early warning systems for dike failures and evacuation plans of people at risk. Regrettably, the significance of modeling the breach growth in dikes was not emphasized until recently, not only in the Netherlands but worldwide. The knowledge of the breach erosion process in dikes is still poor and the state-of-the-art of dike breach modeling technology is far from advanced. So far mainly the breach growth in dikes of granular soils (i.e. soils without cohesion, e.g. sand) was studied, see, for instance, the model developed by Visser (1998). In addition, available prototype as well as experimental data of dike failures, which are of high importance for model calibration and validation, are scarce. In this thesis a mathematical model is developed for the process of breach growth in dikes built of cohesive soil (here named clay-dikes throughout the thesis) and for the flow rate through the breach. The model is based on the mechanism of breach development as observed in various tests in the laboratory and the field. Due to the complexity of the dike breaching process, the present study is restricted to homogeneous clay-dikes. Furthermore, possible effects of protection layers on the surface of the dike are not included, except those of a toe protection on the outer slope. Effects of waves are also not taken into account. It is assumed that the breaching process starts with a small initial breach in the crest of the dike. Five stages are distinguished in the breach development process in claydikes, similar to the breaching process in sand-dikes as described by Visser (1998). In Stage I, erosion occurs along the inner slope and, depending on the flow velocity, possibly also along the dike crest, resulting in a decrease of the width and the height of the dike in the breach. Then in Stages II and III, the dike body in the breach is further eroded through a combination of (1) flow shear erosion, (2) fluidization of the surface of the slope, (3) impinging jet scour of dike foundation and (4) discrete headcut slope mass failure, until at the end of Stage III the dike body in the breach has been washed away completely. In the following Stages IV and V, the breach grows further mainly laterally due to principally flow shear erosion along the side-slopes of the breach and the resulting discrete side-slope instability. The breach growth in vertical direction in these two stages relies mainly on the erodibility of the dike foundation, the presence and, if any, strength of a toe protection on the outer slope of the dike, and the presence and, if any, erodibility of a relatively high foreland. The flow through the breach is decelerated by the rising inner water in the polder in Stage V, consequently also the breach growth. This flow ultimately stops when either the inner and outer water levels equal or the outer water level drops below the breach bottom. Laboratory experiments were conducted in a flume at Delft University of Technology (DUT) to improve the understanding of the physics of the breach erosion process in clay-dikes and to provide data for the model calibration and validation. Altogether five tests were performed, one with a sand-dike, four with clay-dikes constructed with different mixtures of fine sand, silt and clay. Much attention was paid to get proper sand-silt-clay mixtures. In all the tests water levels and flow velocities both upstream and downstream of the dike were measured. The process of dike breaching was recorded by both digital video cameras and digital cameras. The evolution of the dike profile was determined from the videos and photographs. When the clay-dikes were overflowed, generally erosion occurred first at locations close to the toe of the dike. The larger erosion rate at the lower part of the downstream slope of the dike induced steepening of the slope in time. This slope evolved gradually into a headcut. Headcut erosion then played an important role in the breach growth. The cohesiveness of the dike material affected remarkably the breach erosion process: the sand-dike test had a much faster erosion rate than the others, and higher clay proportions in the soil mixtures led to lower erosion rates. The model has been calibrated against the data of two DUT laboratory experiments and two EC IMPACT Project laboratory experiments on breach growth in clay-dikes. For modeling of breach growth in dikes, the key problem is the description of the rate of erosion of the dike by the flow, and more for dikes built of cohesive soil than for those built of non-cohesive soil. Erosion of cohesive soil is a complicated process and its mathematical description is still not satisfactory. The crucial soil erodibility coefficient Me used in existing erosion formulae is often stated as an experimentally or empirically determined constant. Therefore, based on the calibration results of the four laboratory experiments, an expression has been developed for the calculation of Me according to relevant soil properties. With this expression, the validation of the model against the data of the other two DUT laboratory experiments on clay-dike breaching yields reasonable agreement between the model predictions and the measurements. Finally, the model has been confronted with a prototype dike failure in China in 1998. The predicted final breach width of 274 m is about 40% smaller than the observed 390 m. The predicted 5.6 x 108 m3 of diverted floodwater volume is very close to the investigation-based estimation of 5.2 x 108 m3.