Background: Several node staging schemes have been proposed for gastric cancer. The optimal system remains controversial. Methods: Patients with gastric cancer were identified from the Surveillance, Epidemiology, and End Results (SEER) database, and a Chinese patient cohort was used for independent validation. The prognostic performance of three node staging schemes was compared, involving a number-based scheme (pN), ratio-based scheme (rN) and log odds of positive lymph nodes scheme (LODDS). Results: There were 12 443 patients in the SEER database and 866 in the Chinese cohort. LODDS provided better discriminatory capacity and higher predictive accuracy than either pN or rN, for patients with gastric cancer in both the SEER database and the Chinese cohort. The multivariable model using the LODDS classification was significantly more predictive than the pN classification. LODDS suffered much less from stage migration and was able efficiently to discriminate the heterogeneity for patients with no nodes involved or all nodes involved, whereas the pN and rN schemes could not. Conclusion: LODDS showed a clear prognostic superiority over both pN and rN schemes. It could serve as an important reference for the tumour node metastasis (TNM) node classification.