OBJECTIVE: To quantify prediction of gait velocity in ambulatory stroke patients during rehabilitation. DESIGN: Single group (n = 42) at the beginning of rehabilitation (Test 1) and 8 weeks later (Test 2). SETTING: Inpatient rehabilitation. PATIENTS: Unilateral first stroke; informed consent; able to walk 10 meters. MEASURES: INDEPENDENT VARIABLES: Gait velocity at Test 1, age, time from stroke to Test 1, side of lesion, neglect. DEPENDENT VARIABLES: Gait velocity at Test 2, gait velocity change. RESULTS: The correlation between initial gait velocity and gait velocity outcome at Test 2 was of moderate strength (r2 = .62, p<.05). However, even at its lowest, the standard error of prediction for an individual patient was 9.4 m/min, with 95% confidence intervals extending over a range of 36.8 m/min. Age was a weak predictor of gait velocity at Test 2 (r2 = -.10, p<.05). Gait velocity change was poorly predicted. The only significant correlations were initial gait velocity (r2 = .10, p<.05) and age (r2 = .10, p<.05). CONCLUSION: While the prediction of gait velocity at Test 2 was of moderate strength on a group basis, the error surrounding predicted values of gait velocity for a single patient was relatively high, indicating that this simple approach was imprecise on an individual basis. The prediction of gait velocity change was poor. A wide range of change scores was possible for patients, irrespective of their gait velocity score on admission to rehabilitation.