BACKGROUND: We tested the feasibility of delivering and evaluating a complex therapy intervention which aimed to promote activity and independence for people with early dementia (PrAISED). Feasibility questions were on: recruitment, randomisation, intervention delivery, adherence and withdrawals, level of supervision required, adverse events, data collection and sample size assumptions. METHODS: We conducted a three-arm, multi-site, single-blind, randomised controlled feasibility trial. Eligibility criteria were aged 65 years or older, diagnosed mild dementia or mild cognitive impairment, able to walk without human help, and communicate in English, no co-morbidities that prevented participation in cognitive assessment and capacity to give consent. Participants were recruited from Memory Assessment Service clinics and the 'Join Dementia Research' register. Patient participants were randomised 1:1:1 to a high intensity supervision PrAISED intervention, moderate intensity supervision PrAISED intervention or brief falls prevention assessment and advice (control). The PrAISED intervention aimed for participants to complete three hours of PrAISED exercises a week for 12 months. It included individualised activity and exercise plans and supervised exercises with regular re-assessment and progression, and was delivered by occupational therapists, physiotherapists and rehabilitation support workers. Primary efficacy outcome was the Disability Assessment for Dementia (DAD), measured after 12 months. Secondary outcomes included physical activity, quality of life, mood, cognition, strength, balance, rate of falls, frailty and carer strain. Falls and activity were ascertained by monthly diary. RESULTS: Between September 2016 and March 2017 we recruited 60 patient participants and 54 carer participants from two sites. Forty-nine patient participants completed a follow-up interview. Feasibility outcomes were mostly satisfactory, including recruitment and retention, intervention delivery and data completeness for most scales used. We could not maintain blinding of researchers at follow-up and experienced difficulties collecting data using some questionnaires and devices. Participants only completed a mean 77 (moderate supervision) and 71 (high supervision) minutes per week of PrAISED exercises over 12 months. We recorded 19 adverse events, none serious and related to the intervention. CONCLUSION: We conclude that with some adjustments to the trial protocol, it is feasible to deliver the PrAISED intervention and conduct a trial. TRIAL REGISTRATION: ClinicalTrials.gov: NCT02874300 (first posted 22nd August 2016), ISRCTN: 10550694 (date assigned 31st August 2016).