Background: Loss of arm function is a common and distressing consequence of stroke. We describe the protocol for a pragmatic, multicentre randomised controlled trial to determine whether robot-assisted training improves upper limb function following stroke. Methods/design: Study design: a pragmatic, three-arm, multicentre randomised controlled trial, economic analysis and process evaluation. Setting: NHS stroke services. Participants: adults with acute or chronic first-ever stroke (1 week to 5 years post stroke) causing moderate to severe upper limb functional limitation. Randomisation groups: 1. Robot-assisted training using the InMotion robotic gym system for 45 min, three times/week for 12 weeks 2. Enhanced upper limb therapy for 45 min, three times/week for 12 weeks 3. Usual NHS care in accordance with local clinical practice Randomisation: individual participant randomisation stratified by centre, time since stroke, and severity of upper limb impairment. Primary outcome: upper limb function measured by the Action Research Arm Test (ARAT) at 3 months post randomisation. Secondary outcomes: upper limb impairment (Fugl-Meyer Test), activities of daily living (Barthel ADL Index), quality of life (Stroke Impact Scale, EQ-5D-5L), resource use, cost per quality-adjusted life year and adverse events, at 3 and 6 months. Blinding: outcomes are undertaken by blinded assessors. Economic analysis: micro-costing and economic evaluation of interventions compared to usual NHS care. A within-trial analysis, with an economic model will be used to extrapolate longer-term costs and outcomes. Process evaluation: semi-structured interviews with participants and professionals to seek their views and experiences of the rehabilitation that they have received or provided, and factors affecting the implementation of the trial. Sample size: allowing for 10% attrition, 720 participants provide 80% power to detect a 15% difference in successful outcome between each of the treatment pairs. Successful outcome definition: baseline ARAT 0-7 must improve by 3 or more points; baseline ARAT 8-13 improve by 4 or more points; baseline ARAT 14-19 improve by 5 or more points; baseline ARAT 20-39 improve by 6 or more points. Discussion: The results from this trial will determine whether robot-assisted training improves upper limb function post stroke. Trial registration: ISRCTN, identifier: ISRCTN69371850. Registered 4 October 2013.