AIMS: Optimal detection of metastases in sentinel lymph nodes (SLN) remains controversial. To determine the reliability of intraoperative frozen sections, SLN protocol with one frozen section was compared with macroscopic SLN evaluation with consecutive complete SLN embedding. METHODS AND RESULTS: SLN from 135 consecutive breast cancer patients were analysed under a sereomicroscope. Frozen sections were performed in suspicious or clearly involved SLN on cut surface. One control group (n = 143) underwent one intraoperative frozen section on each SLN. The second control group (n = 90) was subjected to stereomicroscopy and one intraoperative frozen section on each SLN. A conventional SLN protocol with cytokeratin immunohistochemistry was performed postoperatively in all cases. All groups were statistically comparable. In the study group metastases were suspected in 21 SLN (16%) under the stereomicroscope and all were confirmed histologically. The negative SLN rate was significantly lower in the study group than in the main control group (47% versus 64%, P = 0.008), suggesting loss of metastases during frozen sections. More macrometastases were detected in the study group (30% versus 15%, P = 0.006); there were no differences in isolated tumour cells or micrometastases. The false-negative rate was significantly lower in the control groups (29% versus 13% and 12%, P = 0.001). CONCLUSIONS: Frozen sections potentially lead to loss or reduced size of metastatic deposits in SLN. Avoiding intraoperative frozen sections on grossly inconspicuous SLN may therefore be justified.