AIM: The traditional architecture based classification system of ductal carcinoma in situ (DCIS) has been criticised on the grounds that individual lesions often show more than one pattern resulting in a large mixed category. New DCIS classification systems have emphasised the importance of cytological grade, which is reputed to be more uniformly expressed throughout a lesion. This study investigates the hypothesis that cytological heterogeneity is less common than architectural heterogeneity within DCIS lesions. METHODS: 121 cases of DCIS were graded as poorly, intermediately, or well differentiated according to a recently developed classification system that employs cytonuclear morphology as the major diagnostic criterion. Cases were categorised as pure when only one grade was present and as mixed if more than one grade was observed. Architecturally the cases were classified as solid, cribriform, micropapillary, or papillary and were described as pure if only one architectural pattern was present and as mixed if more than one pattern was seen. The incidence of cytological heterogeneity was compared with that of architectural heterogeneity. The presence of necrosis was assessed as an independent parameter and the relation to DCIS grade evaluated. RESULTS: Using the cytology based classification system 102 cases (84%) were classified as pure (65 poorly differentiated, 25 intermediately differentiated, and 12 well differentiated) and 19 cases (16%) as mixed. Extensive necrosis was observed in 61 (50%) cases and was closely correlated to DCIS grade. Architecturally 46 cases (38%) were classified as pure (38 solid, 5 cribriform, 2 micropapillary, and 1 papillary) and 75 (62%) as mixed. CONCLUSIONS: Cytological heterogeneity is much less common than architectural heterogeneity in DCIS lesions. The assessment of cytonuclear morphology is therefore likely to provide more consistent information about DCIS, particularly in small biopsy specimens where only part of the lesion may be available for examination.