Abstract Purpose Sentinel lymph node biopsy (SLNB) and axillary lymph node dissection (ALND) are used to assess axillary nodal status in breast cancer, but are invasive procedures associated with morbidity, including lymphoedema. This systematic review evaluates the diagnostic accuracy of positron emission tomography (PET), with or without computed tomography (CT), for assessment of axillary nodes in early breast cancer. Methods Eleven databases including MEDLINE, EMBASE and the Cochrane Library, plus research registers and conference proceedings, were searched in April 2009. Study quality was assessed using the QUality Assessment of Diagnostic Accuracy Studies (QUADAS) checklist. Sensitivity and specificity were meta-analysed using a bivariate random effects approach. Results Across 26 studies evaluating PET or PET/CT ( n  = 2591 patients), mean sensitivity was 63% (95% CI: 52–74%; range 20–100%) and mean specificity 94% (95% CI: 91–96%; range 75–100%). Across 7 studies of PET/CT ( n  = 862), mean sensitivity was 56% (95% CI: 44–67%) and mean specificity 96% (90–99%). Across 19 studies of PET-only ( n  = 1729), mean sensitivity was 66% (50–79%) and mean specificity 93% (89–96%). Mean sensitivity was 11% (5–22%) for micrometastases (≤2 mm; five studies; n  = 63), and 57% (47–66%) for macrometastases (>2 mm; four studies; n  = 111). Conclusions PET had lower sensitivity and specificity than SLNB. Therefore, replacing SLNB with PET would avoid the adverse effects of SLNB, but lead to more false negative patients at risk of recurrence and more false positive patients undergoing unnecessary ALND. The present evidence does not support the routine use of PET or PET-CT for the assessment of the clinically negative axilla.