Background Brain tumour patients see their primary care doctor on average three or more times before diagnosis, so there may be an opportunity to identify ‘at risk’ patients earlier. Suspecting a brain tumour diagnosis is difficult because brain tumour-related symptoms are typically non-specific. Methods We explored the predictive value of referral guidelines (Kernick and NICE 2005) for brain imaging where a tumour is suspected, in a population-based patient group referred for direct access CT of the head. A consensus panel reviewed whether non-tumour findings were clinically important or whether further investigation was necessary. Results Over a 5-year period, 3257 head scans were performed; 318 scans were excluded according to pre-specified criteria. 53 patients (1.8%) were reported to have intracranial tumours, of which 42 were significant (diagnostic yield of 1.43%). There were no false negative CT scans for tumour. With symptom-based referral guidelines primary care doctors can identify patients with a 3% positive predictive value (PPV). 559 patients had non-tumour findings, 31% of which were deemed clinically significant. In 34% of these 559 patients, referral for further imaging and/or specialist assessment from primary care was still thought warranted. Conclusion Existing referral guidelines are insufficient to stratify patients adequately based on their symptoms, according to the likelihood that a tumour will be found on brain imaging. Identification of non-tumour findings may be significant for patients and earlier specialist input into interpretation of these images may be beneficial. Improving guidelines to better identify patients at risk of a brain tumour should be a priority, to improve speed of diagnosis, and reduce unnecessary imaging and costs. Future guidelines may incorporate groups of symptoms, clinical signs and tests to improve the predictive value.