Case mix adjustment models for long-term stroke rehabilitation outcomes should be developed (1) to facilitate equitable comparisons of outcomes across treatment settings, thereby reducing disincentives for treating complex cases, (2) to improve triage into the most appropriate level of rehabilitative care after discharge from acute care, and (3) to confirm that case mix factors are equated in treatment effectiveness studies and by random assignment across conditions in clinical trials. Case mix adjustment is necessary for valid quality improvement processes. A conceptual model of case mix adjustment of long-term rehabilitation outcomes is presented that (1) is diagnosis-specific, (2) includes demographic variables as important case mix factors, (3) encompasses triage into rehabilitation as well as treatment processes as aspects of quality of rehabilitative care, (4) contains outcomes measuring functional status as well as mortality and morbidity, and (5) keys timing of outcomes to onset of conditions requiring rehabilitation rather than discharge from rehabilitation. The number of potential interactions among case mix indicators requires a sophisticated analytic framework. Random factors in the model illustrate that case mix adjustment can never be perfect. Nevertheless, it is essential. A brief review of the stroke literature on prediction of long-term outcomes suggests that additional work is needed to specify relevant case mix indicators.