Interventions that improve cognitive function in Alzheimer's disease are urgently required. To assess whether a novel cognitive training paradigm based on 'chunking' improves working memory and general cognitive function, and is associated with reorganisation of functional activity in prefrontal and parietal cortices (trial registration: ISRCTN43007027). Thirty patients with mild Alzheimer's disease were randomly allocated to receive 18 sessions of 30 min of either adaptive chunking training or an active control intervention over approximately 8 weeks. Pre- and post-intervention functional magnetic resonance imaging (fMRI) scans were also conducted. Adaptive chunking training led to significant improvements in verbal working memory and untrained clinical measures of general cognitive function. Further, fMRI revealed a bilateral reduction in task-related lateral prefrontal and parietal cortex activation in the training group compared with controls. Chunking-based cognitive training is a simple and potentially scalable intervention to improve cognitive function in early Alzheimer's disease.