Transcriptomic analysis of kidney biopsies has demonstrated the potential to improve diagnosis of allograft rejection. Here, we developed a molecular assessment of antibody-mediated rejection (AMR) and T cell-mediated rejection (TCMR) based on the Banff Human Organ Transplant consensus gene panel. Expression assays of formalin-fixed paraffin-embedded kidney biopsies from well-phenotyped cohorts were used to develop prediction models for AMR and TCMR and an automated report of gene expression-based diagnosis. The study population consisted of 950 kidney allograft biopsies from 10 transplantation centers in Europe and North America. The development cohort included 664 renal allograft biopsies split into a training (n = 537) and test set (n = 127), and 2 external validation cohorts (n = 286). We performed gene selection using regularized regression and developed several different base models based on Banff Human Organ Transplant expression data, which were combined into a single ensemble model for each rejection diagnosis. Model performance was assessed in the test set and the 2 external validation cohorts, showing good discriminative abilities (respective areas under the precision-recall curve: AMR = 0.811, 0.891, and 0.832 and TCMR = 0.736, 0.810, and 0.782). We identified challenging biopsies with histology below diagnostic thresholds for which gene expression-based probability can refine rejection diagnosis. This automated molecular diagnostic system shows potential for improving kidney allograft rejection diagnosis in routine practice and clinical trials.