In the past decade, we have witnessed a period of unparallel development in the field of cancer genomics. To address the same or similar biomedical questions, multiple cancer genomic studies have been independently designed and conducted. Cancer gene signatures identified from analysis of individual datasets often have low reproducibility. A cost-effective way of improving reproducibility is to conduct integrative analysis of datasets from multiple studies with comparable designs. To properly integrate multiple studies and conduct integrative analysis, we need to access various public data warehouses, retrieve experiment protocols and raw data, evaluate individual studies and select those with comparable designs, and develop novel statistical methods that can naturally accommodate the heterogeneity among studies and can identify genes with consistent effects across multiple studies. In this article, we discuss new developments and challenges associated with integration and integrative analysis of cancer genomic data. Special attentions are paid to newly developed statistical methods for genomic marker selection in integrative analysis.