A generic model of a rowing craft is presented and used to identify the mechanics of the rowing stroke that give rise to variations in the speed of a rowing shell, together with the causes and locations of maximum and minimum shell speeds through the stroke. The model is implemented computationally in a velocity prediction program which uses readily available measured data for the input and simulates the motion of the athlete using a seven-body-segment model. The rowing model is derived to avoid limitations that arise from the quasi-steady application of steady state data to the unsteady flow around the oar blades and includes hydrodynamic and aerodynamic resistance models. Propulsive drive forces arise as a result of the difference between the gate and handle forces due to the lever action of the oar. The surge motion is dominated by the inertial forces arising as a result of the athlete's motion within the craft which are much larger in magnitude than the drive forces. The minimum surge speed occurs midway through the drive phase. Two distinct maxima in the surge speed occur: one at the end of the drive as the athlete comes to rest at the end of the stroke, and one during the recovery as the athlete's seat attains the maximum speed backwards relative to the craft. The relative magnitudes of these peaks are mainly determined by the ratio of the drive time to the recovery time.