Metastatic non–small-cell lung cancer remains a fatal disease with a median survival of < 1 year. A critical challenge is to develop predictive markers for customizing platinum-based treatment. The first studies focused on the excision repair cross-complementing 1 (ERCC1) gene in this difficult task. Several layers of evidence indicate that ERCC1 mRNA expression could be a predictive marker for cisplatin alone or in combination with certain drugs such as etoposide, gemcitabine, and 5-fluorouracil but not in combination with antimicrotubule drugs. Several retrospective studies demonstrated an impressive survival advantage for gemcitabine plus cisplatin but not for other combinations in tumors with low ERCC1 expression. A customized phase III ERCC1-based trial met the primary endpoint of improvement in response but not in survival, leading us to hypothesize that docetaxel might not be the most appropriate partner for cisplatin in the presence of low ERCC1 levels or for gemcitabine in the presence of high ERCC1 levels. A phase II study demonstrated the feasibility of combining carboplatin, gemcitabine, docetaxel, and vinorelbine according to ERCC1 and ribonucleotide reductase subunit M1 expression levels. These findings highlight the importance of continual learning, and decision-making strategies for customizing treatment should reflect the limitations of our knowledge.