This paper investigates the repetitive learning composite anti-disturbance fault-tolerant (CADFT) control issue for a class of unmanned marine vehicles (UMVs) with unmeasured velocity information, actuator faults and multiple disturbances. To overcome the absence of velocity information, a new integrated-fault-disturbance estimator is proposed to achieve the accurate estimation of the sum of faults and disturbances. Based on the observed information, a new kind of observer-based proportional-derivative (PD)-type repetitive learning CADFT controller is developed to realise the periodic trajectory tracking. The proposed repetitive learning controller avoids using the time derivative of measured outputs and provides more freedom degree of control than traditional PD-type repetitive learning controller. Finally, a numerical simulation on UMVs in the dynamic positioning mode is carried out to show the effectiveness and superiority of the proposed repetitive learning CADFT control method.