The validity of sentinel lymph node biopsy (SLNB) for T1 or T2, clinically N0, oral cancer was tested by correlation of sentinel node pathologic status with that of nodes within the completion neck dissection. This prospective, cooperative group trial involved 25 institutions over a 3-year period. One hundred forty patients with invasive oral cancers, stage T1 and T2, N0 including 95 cancers of the tongue, 26 of the floor of mouth, and 19 other oral cancers were studied. The study excluded lesions with diameter smaller than 6 mm or minimal invasion. Imaging was used to exclude nonpalpable gross nodal disease. Patients underwent injection of the lesion with (99m)Tc-sulfur colloid, nuclear imaging, narrow-exposure SLNB, and completion selective neck dissection. The major end point was the negative-predictive value (NPV) of SLNB. In the 106 SLNBs, which were found to be pathologically and clinically node-negative by routine hematoxylin and eosin stain, 100 patients were found to have no other pathologically positive nodes, corresponding to a NPV of 94%. With additional sectioning and immunohistochemistry, NPV was improved to 96%. In the forty patients with proven cervical metastases, the true-positive rate was 90.2% and was superior for tongue tumors relative to floor of mouth. For T1 lesions, metastases were correctly identified in 100%. For T1 or T2 N0 oral squamous cell carcinoma, SLNB with step sectioning and immunohistochemistry, performed by surgeons of mixed experience levels, correctly predicted a pathologically negative neck in 96% of patients (NPV, 96%).