A novel grading method which utilises intra-cellular mucin content and tubular differentiation (the Goseki grade) has been applied to 181 gastric cancers removed in potentially curative resections, and compared to conventional tumour grading (well, moderate, poor differentiation) and the Lauren, Ming and WHO classifications. The Goseki grade is significantly related to patient survival and, unlike the existing approaches, remains so after allowing for tumour stage in multivariate analyses. Potential relationships between the Goseki grade, direct tumour spread, lymph node involvement and survival have been investigated. Prognosis is particularly related to mucin content; 55% of patients with mucin-rich tumours dying within 5 years compared to 29% of those with mucin-poor cancers. A smaller proportion of mucin-poor cancers show lymph node involvement (N1 and N2) than mucin-rich tumours (52% v. 73%) and a smaller proportion of cancers showing good tubular differentiation exhibit widespread lymph node involvement (N2) than tubule-poor cancers (11% v. 28%). However, these differences did not achieve statistical significance. When compared to the existing classification or grading methods, the Goseki grade was found to be highly significantly correlated with the WHO and Lauren classifications and to conventional grading, but not to the Ming classification. These inter-relations are largely dependent upon tubular differentiation. There was no relationship between Goseki grade and the lymphocytic response around the tumour margin. Inter-observer agreement on the WHO type and the Lauren, Ming and Goseki grading methods was tested on 70 randomly selected cases.(ABSTRACT TRUNCATED AT 250 WORDS)