Brain biopsy has an uncertain role in the diagnosis of dementia. Here we report a retrospective analysis of 90 consecutive cerebral biopsies undertaken for the investigation of dementia in adults at a tertiary referral centre between 1989 and 2003. In most cases (90%), biopsy consisted of a right frontal full thickness resection of cortex, white matter and overlying leptomeninges. Fifty-seven per cent of biopsies were diagnostic: the most frequent diagnoses were Alzheimer's disease (18%), Creutzfeldt-Jakob disease (12%) and inflammatory disorders (9%). Other diagnoses in individual patients included Pick's disease, corticobasal degeneration and other tauopathies, Lewy body dementia, multiple sclerosis, Whipple's disease, progressive multifocal leucoencephalopathy, cerebral autosomal dominant arteriopathy with subcortical ischaemic leucoencephalopathy, vasculopathies and paraneoplastic encephalopathy. The most frequent biopsy finding in the non-diagnostic group and for the series as a whole (37%) was non-specific gliosis variably affecting both cortex and white matter. Complications (11%) included seizures, intracranial and wound infections, and intracranial haemorrhage; there were no deaths or lasting neurological sequelae attributable to the procedure. No trends in diagnostic yield or complication rate over the course of the series were identified. Information obtained at biopsy determined treatment in 11%. A raised cerebrospinal fluid cell count was the only robust predictor of a potentially treatable (inflammatory) process at biopsy. The constellation of behavioural change, raised CSF protein and matched oligoclonal bands in CSF and serum was associated with non-specific gliosis at biopsy. This series underlines the value of cerebral biopsy in the diagnosis of dementia, and suggests that certain clinical and laboratory features may be useful in guiding the decision to proceed to brain biopsy where a treatable disease cannot be excluded by other means.