Background: Several studies have shown that stroke units improve patient outcome under test conditions. The reproducibility of these findings in mainstream practice remains unknown. Patients and Methods: An observational study over 3 years was undertaken on a stroke rehabilitation unit to determine variances in patient composition and outcome from data collected during a randomised controlled study. Results: Three hundred and sixty patients were managed on the stroke rehabilitation unit over the 3-year period of the study. Of these, 124 participated in a randomised study (Year 1) and 236 were managed on the stroke rehabilitation unit in the 2 years after the study (Years 2 and 3). The proportion of patients with very severe strokes managed on the unit increased from 18 (15%) in Year 1 to 33 (27%) in Year 3 (p < 0.01). The outcome in stroke patients with intermediate prognosis managed on the unit after its transition to mainstream practice was similar to the outcome observed under trial conditions. Despite comparable stroke characteristics the number of severe stroke patients being discharged home and median discharge Barthel Index rose significantly over the 3-year period (discharge home: 48 vs. 16%, p < 0.02; discharge Barthel Index: 9 vs. 6, p < 0.05). Conclusions: Mainstream stroke rehabilitation units have a higher proportion of severely disabled stroke patients compared with trial units. Outcome observed in stroke patients within each prognostic group managed on the stroke rehabilitation unit during trial conditions is maintained after transition to mainstream practice.