Short echo time localized proton magnetic resonance spectroscopy provides quantification of brain metabolites, including N-acetyl-aspartate, myo-inositol, creatine/phosphocreatine and choline-containing compounds, which may be useful biomarkers for monitoring Alzheimer's disease. We aimed to quantify the rate of metabolite change in Alzheimer's disease, to assess factors influencing changes and to investigate the potential for serial magnetic resonance spectroscopy as an Alzheimer's disease trial biomarker. A total of 42 patients and 22 controls each had up to six magnetic resonance spectroscopy examinations over a 2-year period, using a midline posterior cingulate single-voxel point resolved spectroscopy sequence (1.5 T; time to repetition = 2000 ms; echo time = 30 ms; 192 averages). Metabolite ratios N-acetyl-aspartate:creatine/phosphocreatine, choline-containing compounds:creatine/phosphocreatine, and myo-inositol:creatine/phosphocreatine were measured using online software (PROBE-Q) and the N-acetyl-aspartate:myo-inositol ratio was derived. Baseline ratios were compared between patients and controls. A linear mixed model was used to quantify longitudinal changes and extended to assess the effect of age, disease severity and baseline use of acetylcholinesterase inhibitors. Patients and controls were matched for age (patients: 68.9 ± 7.2 years; controls: 69.1 ± 6.7 years); 71% of the patients were on acetylcholinesterase inhibitors at baseline; mean Mini-Mental State Examination for patients was 19.4 ± 4.1. A total of 307 spectra were acquired. In cross-sectional analyses, patients were significantly different from controls for N-acetyl-aspartate:creatine/phosphocreatine (11% lower, P < 0.001), N-acetyl-aspartate:myo-inositol (24% lower, P < 0.001), and myo-inositol:creatine/phosphocreatine (17% higher, P < 0.001). After adjustment for N-acetyl-aspartate:myo-inositol, none of the other variables differed significantly. In patients there was significant decline in N-acetyl-aspartate:creatine/phosphocreatine (mean: 2.2%/year; 95% confidence interval: 0.9-3.5) and N-acetyl-aspartate:myo-inositol (mean: 3.7%/year; 95% confidence interval: 1.7-5.7), with no evidence for influence by age, disease severity or acetylcholinesterase inhibitor use. There was significant excess decline in patients compared with controls only in N-acetyl-aspartate:myo-inositol (mean: 3.6%/year; 95% confidence interval: 0.8-6.4; P = 0.014). Between-subject standard deviation for N-acetyl-aspartate:myo-inositol was 0% for controls and 3.5%/year for patients; within-subject standard deviation for a 1 year, two-time-point study was 9.2%/year for both patients and controls. These results confirm that magnetic resonance spectroscopy can be used to quantify excess metabolite decline in Alzheimer's disease, which may provide a useful measure of disease progression. We found no evidence that age, disease severity or acetylcholinesterase inhibitor use influenced rate of decline, although numbers were small. The substantial variability in longitudinal measurements that drives sample size requirements is principally within-subject and technique related: technical developments to reduce this variability may make serial magnetic resonance spectroscopy a viable biomarker in clinical trials for Alzheimer's disease.