AIMS To determine the eventual advantage of the new 1997 TNM as prognosis predictor for gastric cancer patients submitted to an R0 resection and to compare it with two other lymph-node involvement classifications, the 1990 TNM and the Okusa system. METHODS From January 1980 to December 1995, an R0 resection was performed as primary therapy in 275 cases of gastric cancer. These operations consisted of a total or sub-total gastrectomy and of a D2 type lymph-node dissection. Tumour classification was performed according to 1990 and 1997 TNM systems, and to the Okusa lymph node classification. The statistical methods used to evaluate prognostic value were: Kaplan-Meier survival estimates; the log-rank test for univariate analysis; and Cox's model for multivariate analysis. RESULTS The 1990 TNM showed the best stratification power in univariate analysis. In multivariate analysis, the Okusa classification was identified as the best prognostic index (P<0.01). The 1997 TNM showed worse stratification capability than the two other systems. CONCLUSIONS In the present series, the new TNM (1997) did not improve the prognostic stratification of lymph-node involvement. An adequate and universal system for lymph-node stratification is necessary and further validation of these classifications is needed.