SMAD4 mutation was recently implicated in promoting invasion and poor prognosis of pancreatic cancer (PACA) by regulating the tumor immune microenvironment. However, SMAD4-driven immune landscape and clinical significance remain elusive. In this study, we applied the consensus clustering and weighted correlation network analysis (WGCNA) to identify two heterogeneous immune subtypes and immune genes. Combined with SMAD4-driven genes determined by SMAD4 mutation status, a SMAD4-driven immune signature (SDIS) was developed in ICGC-AU2 (microarray data) via machine learning algorithm, and then was validated by RNA-seq data (TCGA, ICGC-AU and ICGC-CA) and microarray data (GSE62452 and GSE85916). The high-risk group displayed a worse prognosis, and multivariate Cox regression indicated that SDIS was an independent prognostic factor. In six cohorts, SDIS also displayed excellent accuracy in predicting prognosis. Moreover, the high-risk group was characterized by higher frequencies of TP53/CDKN2A mutations and SMAD4 deletion, superior immune checkpoint molecules expression and more sensitive to chemotherapy and immunotherapy. Meanwhile, the low-risk group was significantly enriched in metabolism-related pathways and suggested the potential to target tumor metabolism to develop specific drugs. Overall, SDIS could robustly predict prognosis in PACA, which might serve as an attractive platform to further tailor decision-making in chemotherapy and immunotherapy in clinical settings.