The design of trials of interventions intended to slow or arrest the progression of senile dementia of the Alzheimer type must be based on analysis of the natural history of the disease. Using a random coefficients statistical model, we analyzed the natural history of senile dementia of the Alzheimer type in carefully defined subjects with mild disease (n = 68) for periods of up to 10 years. Subject performance was assessed longitudinally on batteries of clinical and psychometric measures. The characteristics of these measures were analyzed relevant to their utility as outcome measures for long-term trials in patients with senile dementia of the Alzheimer type. Estimates were made of sample sizes required to show arrest, and 50% or 25% slowing in the progression of mild disease. We suggest that a clinically relevant global measure, such as the Sum of Boxes of the Clinical Dementia Rating scale, and a performance-based clinical scale or psychometric measure would be appropriate in a 12- or 24-month trial enrolling subjects with mild senile dementia of the Alzheimer type.