A new model of the erosion by water of cohesive soils is developed using physical principles. The theoretical framework which is developed recognises the changing nature of the eroding surface of a soil. Raindrop impact and overland flow are considered to act upon a soil surface so removing soil from the cohesive original (or parent) soil. Once this soil enters the overland flow, either as aggregates or primary particles, it is considered to return to the soil bed, from which it may be re-removed. The development of a deposited layer makes it necessary to distinguish between processes removing sediment from the original soil and those processes removing the deposited layer. This layer, being formed by the relatively gentle action of deposition during the current erosion event, is presumed cohesionless. The physical properties of the original soil and the deposited layer are considered to be very different. The development of two experimental apparatus, a rainfall/runoff simulator and a settling tube for the measurement of aggregate settling velocities, is first described. Experimental investigations, using these apparatus, and field observations to inform the description of the erosion and deposition processes, are then presented. The processes by which rainfall impact removes sediment from the original soil and the deposited layer are termed rainfall detachment and rainfall re-detachment respectively. Initially, descriptions of these processes in the presence of deposition, are combined in a model describing net rainfall detachment when removal of sediment from the flow bed by overland flow is not occurring. The developriient of the deposited layer is considered both quantitatively and qualitatively. The solution of the equation describing mass conservation is then given for the equilibrium situation when the mass of the deposited layer, and therefore the sediment concentration, is constant with respect to time. The processes by which overland flow removes sediment from the original soil and the deposited layer are termed entrainment and re-entrainment. The work done by the process of entrainment is considered to be done wholly against the cohesive strength of the original soil. In contrast to the process of entrainment, the work done in re-entraining sediment from the deposited layer is considered only to be done against gravity. The resulting description of these processes is then combined with the previous descriptions of rainfall detachment, rainfall re-detachment and deposition and with the equation describing the conservation of mass of sediment within any arbitary number of size (or settling velocity) classes. A plane geometry model Is developed in which the surface water flow is considered to be uniformily distributed across a plane slope on which all processes act. When the mass of the deposited layer is steady, two possible forms of equilibrium are shown to exist. When the coverage of the original soil by deposited layer is partial, the sediment concentration is limited by the removal of the cohesive original soil by entrainment and rainfall detachment, in the presence of deposition. This situation is termed 'source limiting' and is shown to provide a lower limit to sediment concentration. When the coverage of the deposited layer is complete so that entrainment and rainfall detachment of the original soil are considered not to occur, then the ability of the erosive agents to re-entrain and re-detach sediment in the presence of deposition limits sediment concentration. This situation, termed 'transport limiting', is shown to provide a practical upper limit to sediment concentration. This plane geometry flow model is followed by a revised model in which all processes are considered to occur but the flow of water on a plane surface is modified by the formation of rills. In this 'detailed geometry model' the spatial distribution of the erosive agents is shown to have a marked influence on the resulting processes and sediment concentrations. A potential description of the sediment transport across a change in land slope is also developed. Finally, a discussion of this new modelling approach is presented in which the conceptual developments of this thesis are considered and future developments are suggested. This discussion also includes a comparison of the outcomes of this new work with similar erosion models.