Background We aimed to assess the clinical validity of circulating tumour cell (CTC) quantification for prognostication of patients with metastatic breast cancer by undertaking a pooled analysis of individual patient data. Methods We contacted 51 European centres and asked them to provide reported and unreported anonymised data for individual patients with metastatic breast cancer who participated in studies between January, 2003, and July, 2012. Eligible studies had participants starting a new line of therapy, data for progression-free survival or overall survival, or both, and CTC quantification by the CellSearch method at baseline (before start of new treatment). We used Cox regression models, stratified by study, to establish the association between CTC count and progression-free survival and overall survival. We used the landmark method to assess the prognostic value of CTC and serum marker changes during treatment. We assessed the added value of CTCs or serum markers to prognostic clinicopathological models in a resampling procedure using likelihood ratio (LR) χ2 statistics. Findings 17 centres provided data for 1944 eligible patients from 20 studies. 911 patients (46·9%) had a CTC count of 5 per 7·5 mL or higher at baseline, which was associated with decreased progression-free survival (hazard ratio [HR] 1·92, 95% CI 1·73–2·14, p<0·0001) and overall survival (HR 2·78, 95% CI 2·42–3·19, p<0·0001) compared with patients with a CTC count of less than 5 per 7·5 mL at baseline. Increased CTC counts 3–5 weeks after start of treatment, adjusted for CTC count at baseline, were associated with shortened progression-free survival (HR 1·85, 95% CI 1·48–2·32, p<0·0001) and overall survival (HR 2·26, 95% CI 1·68–3·03) as were increased CTC counts after 6–8 weeks (progression-free survival HR 2·20, 95% CI 1·66–2·90, p<0·0001; overall survival HR 2·91, 95% CI 2·01–4·23, p<0·0001). Survival prediction was significantly improved by addition of baseline CTC count to the clinicopathological models (progression-free survival LR 38·4, 95% CI 21·9–60·3, p<0·0001; overall survival LR 64·9, 95% CI 41·3–93·4, p<0·0001). This model was further improved by addition of CTC change at 3–5 weeks (progression-free survival LR 8·2, 95% CI 0·78–20·4, p=0·004; overall survival LR 11·5, 95% CI 2·6–25·1, p=0·0007) and at 6–8 weeks (progression-free survival LR 15·3, 95% CI 5·2–28·3; overall survival LR 14·6, 95% CI 4·0–30·6; both p<0·0001). Carcinoembryonic antigen and cancer antigen 15-3 concentrations at baseline and during therapy did not add significant information to the best baseline model. Interpretation These data confirm the independent prognostic effect of CTC count on progression-free survival and overall survival. CTC count also improves the prognostication of metastatic breast cancer when added to full clinicopathological predictive models, whereas serum tumour markers do not. Funding Janssen Diagnostics, the Nuovo-Soldati foundation for cancer research.