This dissertation studies the large-scale multiple testing problem from a compound decision theoretical view, and proposes a new class of powerful data-driven procedures that substantially outperform the traditional p -value based approaches. There are several important implications from my...