Purpose: The majority of gastric cancer patients are diagnosed with late-stage disease, for which distinct molecular subtypes have been identified that are potentially amenable to targeted therapies. However, there exists no molecular classification system with prognostic power for early-stage gastric cancer (EGC) because the molecular events promoting gastric cancer initiation remain ill-defined. Experimental Design: miRNA microarrays were performed on gastric tissue from the gp130F/F preclinical EGC mouse model, prior to tumor initiation. Computation prediction algorithms were performed on multiple data sets and independent gastric cancer patient cohorts. Quantitative real-time PCR expression profiling was undertaken in gp130F/F-based mouse strains and human gastric cancer cells genetically engineered for suppressed activation of the oncogenic latent transcription factor STAT3. Human gastric cancer cells with modulated expression of the miR-200 family member miR-429 were also assessed for their proliferative response. Results: Increased expression of miR-200 family members is associated with both tumor initiation in a STAT3-dependent manner in gp130F/F mice and EGC (i.e., stage IA) in patient cohorts. Overexpression of miR-429 also elicited contrasting pro- and antiproliferative responses in human gastric cancer cells depending on their cellular histologic subtype. We also identified a miR-200 family–regulated 15-gene signature that integrates multiple key current indicators of EGC, namely tumor invasion depth, differentiation, histology, and stage, and provides superior predictive power for overall survival compared with each EGC indicator alone. Conclusions: Collectively, our discovery of a STAT3-regulated, miR-200 family–associated gene signature specific for EGC, with predictive power, provides a molecular rationale to classify and stratify EGC patients for endoscopic treatment.