Cerebral atrophy calculated from serial MRI is a marker of Alzheimer's disease (AD) progression, and a potential outcome measure for therapeutic trials. Reducing within-subject variability in cerebral atrophy rates by acquiring more than two serial scans could allow for shorter clinical trials requiring smaller patient numbers. Forty-six patients with AD and 23 controls each had up to 10 serial MR brain scans over two years. Whole brain atrophy was calculated for each subject from every scan-pair. 708 volumetric MRI scans were acquired: 2199 measures of atrophy were made for patients, and 1182 for controls. A linear mixed model was used to characterise between and within-individual variability. These results were used to investigate the power of combining multiple serial scans in treatment trials of varying lengths. In AD, the mean whole brain atrophy rate was 2.23%/year (95% CI: 1.90-2.56%/year). The linear mixed model was shown to fit the data well and led to a formula (0.99(2) + (0.82/t)2) for the variance of atrophy rates calculated from two scans "t" years apart. Utilising five optimally timed scans with repeat scans at each visit reduced the component of atrophy rate variance attributable to within-subject variability by approximately 56%, equating to a approximately 40% sample size reduction (228 vs 387 patients per arm to detect 20% reduction in atrophy rate) in a six-month placebo-controlled trial. This benefit in terms of sample size is relatively reduced in longer trials, although adding extra scanning visits may have benefits when patient drop-outs are accounted for. We conclude that sample sizes required in short interval therapeutic trials using cerebral atrophy as an outcome measure may be reduced if multiple serial MRI is performed.