Background: Treatment of patients with malignant large bowel polyps is highly dependent on pathological evaluation. The aim of this study was to evaluate interobserver variability in the pathological assessment of endoscopically removed polyps. Methods: The records of 88 patients with colorectal cancer who underwent endoscopic removal of malignant polyps were reviewed. Study investigators reviewed the initial pathology report; three experienced gastrointestinal pathologists reviewed all slides in a blinded fashion. Interobserver variability of pathological assessment of malignant polyps was analysed by κ statistics. Results: Seventy-six (86 per cent) of the 88 patients had malignant polyps and 12 (14 per cent) had carcinoma in situ. Agreement between experienced pathologists was substantial with regard to T stage (κ = 0·725), resection margin status (κ = 0·668) and Haggitt's classification (κ = 0·682), but comparison of initial and experienced pathologists' assessment demonstrated only moderate agreement in these areas (κ = 0·516, κ= 0·555 and κ= 0·578 respectively). Agreement between even experienced pathologists was poor with respect to histological grade of differentiated adenocarcinomas (κ = 0·163) and angiolymphatic vessel invasion (κ = -0·017). Conclusion: Pathological assessment of malignant polyps varies between observers. Specialist pathologists appear to have a higher degree of consensus among themselves than with generalist pathologists with respect to T stage. The high interobserver variability with regard to histological grade of differentiated tumours is clinically irrelevant. However, variability in the assessment of angiolymphatic vessel invasion limits the value of this measurement for clinical decision making.