BACKGROUND This retrospective study included 914 patients who underwent a lymph node dissection at our institute between 1980 and 1985. The primary tumor sites were oral cavity, 287; hypopharynx, 249; larynx, 247; and oropharynx, 131. METHODS On the basis of anatomic considerations, the sentinel nodes for well-lateralized oral cavity tumors were defined as homolateral levels I, II, and III; for oropharyngeal, hypopharyngeal, and laryngeal tumors, the sentinel nodes were defined as levels II and III. We took into account the ipsilateral side of the neck for well-lateralized tumors, and both sides for medium or large tumors. For clinically positive nodes of more than 3 cm, a radical neck dissection was performed. Other patients underwent a selective neck dissection on sentinel nodes, with immediate pathologic evaluation. Modified radical neck dissections with contralateral selective dissection were performed when frozen sections were positive. Patients with positive nodes were given postoperative radiotherapy. RESULTS The prognostic factors studied, using the Cox survival model adjusted on the primary tumor site, surprisingly showed a nonsignificant value for extracapsular spread (P = 0.09), and a significant value for the number of positive nodes (P < 0.001) and for the positive node in or out of the sentinel node sites (P < 0.001). Although the node location factor can be used instead of positive node in or out of the sentinel node site, it has a less significant prognostic value. CONCLUSIONS The most significant prognostic factors are the site of the positive node in or out of the sentinel node and the number of positive nodes; and a more accurate approach can be obtained by combining both factors. Node location in the upper or lower neck remains a substitute prognostic factor for the site of the positive node in or out of the sentinel node.