<i>Background/Aims:</i> The status of imaging findings in the clinical diagnosis of frontotemporal dementia (FTD) remains uncertain; while they may be supportive of a diagnosis of frontotemporal dementia, they are not mandatory. Our aim was to assess patterns of lobar atrophy in a large sample of clinically defined, prospectively studied, patients using a magnetic resonance image (MRI) rating scale, to (1) determine whether imaging findings warrant a more prominent position in FTD diagnosis and (2) correlate the extent of lobar atrophy with clinical data. <i>Methods:</i> We adapted a recently devised post mortem rating scale for FTD to rate lobar atrophy on MRI scans. The areas rated included the frontal cortex and both anterior and posterior temporal regions bilaterally. All available brain scans from all patients seen in the Cambridge Dementia Clinic (n = 258) diagnosed as having FTD, together with controls (n = 20), were used to assess the reliability of the method. A subset of these (n = 121) were used for clinico-anatomic analysis. <i>Results:</i> The scale proved quick and reliable (intra-, inter-rater k = 0.80, 0.67). MRI scans were abnormal in the majority of patients (75%), with focal atrophy present in 100% of semantic dementia (SD) patients. By contrast, nearly half (47%) of the patients with clinical behavioural variant FTD had scans within the normal range. Behavioural cases with normal scans generally had fewer cognitive deficits and milder functional impairment than those with abnormal scans, yet displayed a clinically indistinguishable behavioural syndrome. They were not, however, simply at an earlier stage of the disease. <i>Conclusions:</i> MRI findings should form part of the diagnostic criteria for SD; the absence of atrophy on MRI in many behavioural cases raises the prospect that the behavioural syndrome of FTD is not specific for patients with a neurodegenerative disease.