Tremor is a rapid involuntary movement often seen in patients with neurological conditions such as Multiple Sclerosis (MS) and Parkinson’s disease. This debilitating oscillation can be suppressed by applying functional electrical stimulation (FES) within a closed-loop control system. However, conventional implementations use classical control methods and have proved capable of only limited performance. This paper establishes the feasibility of embedding repetitive control (RC) action to exploit the capability of learning from experience to completely suppress tremor at the wrist via FES regulated co-contraction of wrist extensors/flexors. A nonlinear model structure and associated identification procedure is first proposed to guarantee stability and performance of the RC system. Then a linearising control approach is developed to facilitate transparent RC design, together with a mechanism to preserve patients’ voluntary intention. Experimental evaluation is performed with both unimpaired and neurologically impaired participants using a validated wristrig. For the former group a novel electromechanical system is employed to induce tremor artificially. Results are benchmarked against a well-known classical filtering technique to establish the efficacy of the RC approach. These confirm that the proposed control system with the developed model identification procedure can increase tremor suppression by 43.3% compared with conventional filtering. In addition, the mechanism decreases the interference of RC action with voluntary motion by 20.2% compared with conventional filtering.