Rejection-associated gene expression has been characterized in renal allograft biopsies for cause. The aim is to evaluate rejection gene expression in subclinical rejection and in biopsies with borderline changes or interstitial fibrosis and tubular atrophy (IFTA). We included 96 biopsies. Most differentially expressed genes between normal surveillance biopsies (n = 17) and clinical rejection (n = 12) were obtained. A rejection-associated gene (RAG) score was defined as its geometric mean. The following groups were considered: (a) subclinical rejection (REJ-S, n = 6); (b) borderline changes in biopsies for cause (BL-C, n = 13); (c) borderline changes in surveillance biopsies (BL-S, n = 12); (d) IFTA in biopsies for cause (IFTA-C, n = 20); and (e) IFTA in surveillance biopsies (IFTA-S, n = 16). The outcome variable was death-censored graft loss or glomerular filtration rate decline ≥ 30 % at 2 years. A RAG score containing 109 genes derived from normal and clinical rejection (area under the curve, AUC = 1) was employed to classify the study groups. A positive RAG score was observed in 83% REJ-S, 38% BL-C, 17% BL-S, 25% IFTA-C, and 5% IFTA-S. A positive RAG score was an independent predictor of graft outcome from histological diagnosis (hazard ratio: 3.5 and 95% confidence interval: 1.1-10.9; p = 0.031). A positive RAG score predicts graft outcome in surveillance and for cause biopsies with a less severe phenotype than clinical rejectio