Clinical features at the initial examination of 42 patients with probable Alzheimer's disease were tested for prognostic value at subsequent follow-up of 54 +/- 25 months. These potential prognostic features were of three types: degree of severity features (eg, IQ scores); variable clinical features (eg, extrapyramidal signs); and individual distinguishing features (eg, gender, education, and age). The power of these potential prognostic features to predict prognosis was assessed using the Kaplan-Meier life-tables method and the Cox proportional hazards model. Three clinical end points were considered: total dependence in activities of daily living; incontinence; and institutionalization at follow-up. Degree of severity features (subtests of the Wechsler Adult Intelligence Scale-Revised and the Wechsler Memory Scale, and the Clinical Severity Score) predicted subsequent dependence in activities of daily living, incontinence, and institutionalization. Historical disease duration, age, gender, family history of dementia, retrospective rate of progression, anxiety, psychosis, depression, and extrapyramidal signs did not influence prognosis. These results suggest that initial degree of severity ("how far") rather than variation in the rate of progression ("how fast") best predicts prognosis in the early to intermediate stages of Alzheimer's disease. The relationship of disease severity to prognosis should be taken into account before concluding that there are subtypes of Alzheimer's disease that have different rates of progression.