Current assessments do not provide reliable factors predictive of outcome from stroke for stroke survivors of intermediate age and severity of deficit. We sought to investigate whether early rate of functional improvement can facilitate prediction of functional outcome, length of stay, and disposition beyond that afforded by age and initial severity of deficit. Prospective study of consecutive admissions to acute rehabilitation (N = 244) with diagnosis of ischemic or hemorrhagic stroke. Independent measures were age, marital status, living situation, social situation, lag from symptom onset to rehabilitation, stroke type, admission score on the Functional Independence Measure (FIM), rate of FIM change (ROFC) as assessed by the best weekly FIM change in the first 3 weeks of rehabilitation. Dependent measures were functional status on discharge as assessed by a modification of Steinman's method, length of stay, and discharge disposition. Logistic regression analyses on each of the dependent measures identified significant factors, and interactions of significant factors were assessed by analysis of variance on continuous dependent variables. Cross-tabulations using significant factors from the logistic regression analyses were performed to identify groups with homogeneous outcomes. Groups with >80% homogeneity were considered likely to have predictive value. Discharge functional status: Admission FIM (AFIM) again fractionated the population into groups with poor outcome (AFIM <50 remained dependent), good outcome (AFIM >70 achieved nondependence), and an intermediate group with unpredictable outcome. In this intermediate group, ROFC had significant effect only for a small number of patients (n = 9) with rapid improvement (ROFC >25) who achieved nondependence. LOS: AFIM >70 had less than average LOS, ROFC = 10-15 FIM/week had longer than average LOS. LOS was significantly prolonged in patients with poor outcomes. Disposition: AFIM >70 and age <60 were strongly associated with home discharge. Patients not living at home before admission were not discharged home. Married patients had a greater tendency to home discharge than did those not married. ROFC had no bearing on disposition. ROFC has an independent influence on outcome but was sufficiently powerful in our sample to identify reliably only a very small subset of patients with otherwise indeterminate prognosis. LOS seems inordinately prolonged in patients with poor outcomes. Both of these results can guide efficient rehabilitation management.