From March 1996 to December 1999 we performed 1,266 sentinel node biopsies (SNBs) in patients with small breast cancers. The technique is to inject technetium 99m-labeled albumin particles close to the tumor, locate the sentinel node (SN) scintigraphically, and use a handheld gamma-detecting probe to guide its removal via a small incision during breast surgery. Our experience was divided into three phases. In the first phase, complete axillary dissection was performed to assess the accuracy of SNB in predicting axillary status. We also assessed safety, perfected tracer injection technique, determined optimal particle size and radioactivity levels, optimized lymphoscintigraphic scanning, and perfected the surgical technique. The SN was identified and removed in 98.7% of cases. Comparison with complete axillary dissection showed that the SN predicted axillary status in 96.8% of cases. However, use of an intraoperative frozen section method predicted axillary status in only 86.5% of cases. In the second phase we developed a new method for intraoperative histologic analysis. Extensive sampling (up to 60 sections/SN) and an experienced pathologist proved more important than use of antikeratin immunostaining in identifying tumor cells, and the new method has the accuracy of a definitive histologic examination. The third phase, a randomized trial, closed at the end of 1999. Trial objectives were to confirm that the SN predicts axillary status, to determine the number of axillary relapses, and to assess overall and disease-free survival. Patients were randomized in the operating room to complete axillary dissection or SNB. If the SN was positive, complete axillary dissection was performed; if the SN was negative, no further axillary treatment was given. We expect the trial to confirm our clinical experience that SNB is a safe and accurate procedure for staging patients with early breast cancer and a clinically negative axilla.