INTRODUCTION: We aimed to assess the value of a second MR scan in the radiological diagnosis of dementia. METHODS: One hundred twenty subjects with clinical follow-up of at least 1 year with two scans were selected from a cognitive disorders clinic. Scans were reviewed as a single first scan (method A), two unregistered scans presented side-by-side (method B) and a registered pair (method C). Scans were presented to two neuroradiologists and a clinician together with approximate scan interval (if applicable) and age. Raters decided on a main and subtype diagnosis. RESULTS: There was no evidence that differences between methods (expressed as relative odds of a correct response) differed between reviewers (p = 0.17 for degenerative condition or not, p = 0.5 for main diagnosis, p = 0.16 for subtype). Accordingly, results were pooled over reviewers. For distinguishing normal/non-progressors from degenerative conditions, the proportions correctly diagnosed were higher with methods B and C than with A (p = 0.001, both tests). The difference between method B and C was not statistically significant (p = 0.18). For main diagnosis, the proportion of correct diagnoses were highest with method C for all three reviewers; however, this was not statistically significant comparing with method A (p = 0.23) or with method B (p = 0.16). For subtype diagnosis, there was some evidence that method C was better than method A (p = 0.01) and B (p = 0.048). CONCLUSIONS: Serial MRI and registration may improve visual diagnosis in dementia.