The study's objective was to determine the effectiveness of a joint NHS/Social Services rehabilitation unit (a form of intermediate care) for older people on discharge from community hospital, compared with 'usual' community services. This was a controlled clinical trial in a practice setting. The intervention was 6 weeks in a rehabilitation unit where individuals worked with care/rehabilitation assistants and occupational therapists to regain independence. Controls went home with the health/social care services they would ordinarily receive. Participants were from two matched geographical areas in Devon: one with a rehabilitation unit, one without. Recruitment was from January 1999 to October 2001 in 10 community hospitals. Study eligibility was assessed using the unit's inclusion/exclusion criteria: 55 years or older and 'likely to benefit from a short-term rehabilitation programme' ('potential to improve', 'realistic and achievable goals' and 'motivation to participate'). Ninety-four people were recruited to the intervention and 112 to the control. The mean (standard deviation) age was 81.8 (8.0) years. The main outcome measure was prevention of institutionalisation assessed by the number of days from baseline interview to admission to residential/nursing care or death ('survival-at-home time'). Secondary outcome measures were time to hospital re-admission over 12 months, quality of life and coping ability. There were no significant differences between the groups on any outcome measure. Adjusted hazard ratio (95% CI) for 'survival-at-home time' was 1.13 (0.70-1.84), and 0.84 (0.53-1.33) for 'time to hospital re-admission'. However, attending the unit was associated with earlier hospital discharge. Median (interquartile range) days in hospital for the intervention graph was 27 (20, 40), and for the control graph was 35 (22, 47) (U = 4234, P = 0.029). These findings suggest a stay in a rehabilitation unit is no more effective than 'usual' care at diverting older people from hospital/long-term care. Alternative service configurations may be as effective, having implications for tailoring services more specifically to individual need and/or user preferences. However, the unit did appear to facilitate earlier discharges from community hospital.