Purpose : A significant proportion of acute myocardial infarction (MI) patients develop heart failure (HF). Early identification of patients at risk of developing HF after MI would be a major breakthrough. An approach combining the power of biological information networks and the precision of microarray analysis was undertaken to identify new biomarkers of HF. Methods : Since angiogenesis may be related to MI and HF, a protein-protein interaction (PPI) network was constructed by first extracting from the Entrez-Gene database a set of genes relevant to angiogenesis and MI. These genes were used as inputs to retrieve annotated interactions from the Human Protein Reference Database. Potential biomarkers were identified by network analysis. Gene expression profiles of blood cells taken at the time of MI in two groups of 16 patients (high ejection fraction (EF) at 1 month, EF≥45% and low EF at 1 month, EF≤40%) were obtained using oligonucleotide microarrays containing 25,000 genes and compared by Statistical Analysis of Microarrays (SAM). Prediction models based on machine learning were used to classify low and high EF patients. Results : SAM identified 525 genes differentially expressed between patients with high and low EF (fold-change ≥1.3). The PPI network included 556 nodes (proteins) and 686 edges (interactions). A network clustering algorithm identified 53 proteins highly specialized in growth and regulation processes. Out of these, 38 were found differentially expressed by SAM. Further filtering reported 3 genes as the optimal biomarker set: Vascular Endothelial Growth Factor B (VEGFB), Placental Growth Factor (PGF), both pro-angiogenic, and the anti-angiogenic protein Thrombospondin-1 (THBS1). Prediction models reported areas under the receiver operating characteristic curve (AUC) of 0.82 for this biomarker set. Conclusion : The classification performances achieved with the 3 biomarkers stresses the prognostic value of genes involved in angiogenesis. The network-based approach allowed us to identify powerful biomarkers, which could not be identified by applying standard gene expression data analysis only. Therefore, combined network and microarray analysis allows a systematic and less biased approach to biomarker discovery.