OBJECTIVES: To re-examine the evidence for recommendations for complete dissection versus sampling of ipsilateral mediastinal lymph nodes during lobectomy for cancer. METHODS: We searched for randomized trials of systematic mediastinal lymphadenectomy versus mediastinal sampling. We performed a textual analysis of the authors' own starting assumptions and conclusion. We analysed the trial designs and risk of bias. We extracted data on early mortality, perioperative complications, overall survival, local recurrence and distant recurrence for meta-analysis. RESULTS: We found five randomized controlled trials recruiting 1980 patients spanning 1989-2007. The expressed starting position in 3/5 studies was a conviction that systematic dissection was effective. Long-term survival was better with lymphadenectomy compared with sampling (Hazard Ratio 0.78; 95% CI 0.69-0.89) as was perioperative survival (Odds Ratio 0.59; 95% CI 0.25-1.36, non-significant). But there was an overall high risk of bias and a lack of intention to treat analysis. There were higher rates (non-significant) of perioperative complications including bleeding, chylothorax and recurrent nerve palsy with lymphadenectomy. CONCLUSIONS: The high risk of bias in these trials makes the overall conclusion insecure. The finding of clinically important surgically related morbidities but lower perioperative mortality with lymphadenectomy seems inconsistent. The multiple variables in patients, cancers and available treatments suggest that large pragmatic multicentre trials, testing currently available strategies, are the best way to find out which are more effective. The number of patients affected with lung cancer makes trials feasible.