Objective Endoscopic surveillance for Barrett's oesophagus (BO) is limited by sampling error and the subjectivity of diagnosing dysplasia. We aimed to compare a biomarker panel on minimal biopsies directed by autofluorescence imaging (AFI) with the standard surveillance protocol to derive an objective tool for dysplasia assessment. Design We performed a cross-sectional prospective study in three tertiary referral centres. Patients with BO underwent high-resolution endoscopy followed by AFI-targeted biopsies. 157 patients completed the biopsy protocol. Aneuploidy/tetraploidy; 9p and 17p loss of heterozygosity; RUNX3, HPP1 and p16 methylation; p53 and cyclin A immunohistochemistry were assessed. Bootstrap resampling was used to select the best diagnostic biomarker panel for high-grade dysplasia (HGD) and early cancer (EC). This panel was validated in an independent cohort of 46 patients. Results Aneuploidy, p53 immunohistochemistry and cyclin A had the strongest association with dysplasia in the per-biopsy analysis and, as a panel, had an area under the receiver operating characteristic curve of 0.97 (95% CI 0.95 to 0.99) for diagnosing HGD/EC. The diagnostic accuracy for HGD/EC of the three-biomarker panel from AFI+ areas was superior to AFI??? areas (p