Objectives: To (1) determine which clinical assessments at admission to an inpatient rehabilitation facility (IRF) most simply predict discharge walking ability, and (2) identify a clinical decision rule to differentiate household versus community ambulators at discharge from an IRF. Design: Retrospective cohort study. Setting: IRF. Participants: Two samples of participants (n=110 and 159) admitted with stroke. Interventions: A multiple regression determined which variables obtained at admission (age, time from stroke to assessment, Motricity Index, somatosensation, Modified Ashworth Scale, FIM, Berg Balance Scale, 10-m walk speed) could most simply predict discharge walking ability (10-m walk speed). A logistic regression determined the likelihood of a participant achieving household (<0.4m/s) versus community (<0.4-0.8m/s; >0.8m/s) ambulation at the time of discharge. Validity of the results was evaluated on a second sample of participants. Main Outcome Measure: Discharge 10-m walk speed. Results: Admission Berg Balance Scale and FIM walk item scores explained most of the variance in discharge walk speed. The odds ratio of achieving only household ambulation at discharge was 20 (95% confidence interval [CI], 6-63) for sample 1 and 32 (95% CI, 10-96) for sample 2 when the combination of having a Berg Balance Scale score of ≤20 and a FIM walk item score of 1 or 2 was present. Conclusions: A Berg Balance Scale score of ≤20 and a FIM walk item score of 1 or 2 at admission indicates that a person with stroke is highly likely to only achieve household ambulation speeds at discharge from an IRF.