Objective: To evaluate the effects of tai chi consisting of group and home-based sessions in elderly subjects with knee osteoarthritis. Design: A randomized, controlled, single-blinded 12-week trial with stratification by age and sex, and six weeks of follow-up. Setting: General community. Participants: Forty-one adults (70±9.2 years) with knee osteoarthritis. Interventions: The tai chi programme featured six weeks of group tai chi sessions, 40 min/session, three times a week, followed by another six weeks (weeks 7-12) of home-based tai chi training. Subjects were requested to discontinue tai chi training during a six-week follow-up detraining period (weeks 13-18). Subjects in the attention control group attended six weeks of health lectures following the same schedule as the group-based tai chi intervention (weeks 0-6), followed by 12 weeks of no activity (weeks 7-18). Main outcome measures: Knee pain measured by visual analogue scale, knee range of motion and physical function measured by Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) were recorded at baseline and every three weeks throughout the 18-week study period. Data were analysed using a mixed model ANOVA. Results: The six weeks of group tai chi followed by another six weeks of home tai chi training showed significant improvements in mean overall knee pain (P = 0.0078), maximum knee pain (P = 0.0035) and the WOMAC subscales of physical function (P = 0.0075) and stiffness (P = 0.0206) compared to the baseline. No significant change of any outcome measure was noted in the attention control group throughout the study. The tai chi group reported lower overall pain and better WOMAC physical function than the attention control group at weeks 9 and 12. All improvements disappeared after detraining. Conclusions: The group and home based tai chi programme provided significant knee pain reduction and physical function improvement in elderly subjects with knee osteoarthritis. These effects were not sustained after detraining.