MicroRNAs (miRNAs) are an abundant class of small non-coding RNA molecules that regulate gene expression at the post-transcriptional level. MiRNAs are found frequently dysregulated during cancer initiation, development, and metastasis, and are present in a wide variety of clinical specimens such as blood, saliva, urine, and feces. These relatively abundant and stable molecules provide great potential to be exploited for cancer detection, prognosis, and therapy response prediction, as well as disease monitoring. Herein, we introduce state-of-the-art development of miRNA biomarkers with a particular focus on a genome-wide, data-driven methodology, which has demonstrated higher robustness and reproducibility compared to traditional methods. We will first review miRNA-based biomarkers for various clinical applications and discuss the potential limitations of traditional approaches. Next, we will summarize the major steps involved in a data-driven methodology for biomarker development. Finally, we will discuss the main advantages and challenges in real clinical applications, as well as possible solutions and emerging opportunities.