The recently published AJCC-TNM staging system for esophageal carcinoma made an obvious modification on N-classification based on the number of metastatic regional lymph nodes (LN). However, this classification might ignore the site at which these LNs occur, a factor that might be even more important in reflecting patients' prognosis. A retrospective study of 236 patients with carcinoma of thoracic esophagus who underwent esophagectomy between 1984 and 1989 with each at least six LNs removed was conducted, with a 10-year follow-up rate of 92.4%. The proposed scheme for N-classification according to the number (0, 1-2, 3-6, a parts per thousand yen7; N(0-3)), distance (0, 1, 2, 3 stations; S(0-3)), or extent (0, 1, and 2 fields; F(0-2)) of LN involvement was evaluated by univariate and multivariate survival analysis. The LN metastasis was identified in 112 patients, revealing a poorer 5-year survival in this patient group when compared to patients without node involvement. Cox regression analysis revealed that the number and distance of LN metastases and the number of metastasis fields were factors significantly influencing survival. When these factors were further analyzed by univariate log-rank test, no significant difference in survival existed between N(2) and N(3) patients, or among S(1), S(2), and S(3) patients. When patients were grouped according to the extent of LN metastasis, significant differences in survival were observed overall and between each subgroup. Refining the current N-classification for esophageal cancer according to the extent of LN metastasis, rather than by number alone, might be a better means of staging that could subgroup patients more effectively and result in different rates of survival.