Murine models are an essential tool to study human immune responses and related diseases. However, the use of traditional murine models has been challenged by recent systemic surveys that show discordance between human and model immune responses in their gene expression. This is a significant problem in translational biomedical research for human immunity. Here, we describe evidence-based translation (EBT) to improve the analysis of genomic responses of murine models in the translation to human immune responses. Based on evidences from prior experiments, EBT introduces pseudo variances, penalizes gene expression changes in a model experiment, and finally detects false positive translations of model genomic responses that poorly correlate with human responses. Demonstrated over multiple data sets, EBT significantly improves the agreement of overall responses (up to 56%), experiment-specific responses (up to 143%), and enriched biological contexts (up to 100%) between human and model systems. In addition, we provide the category of genes specifically benefiting from EBT and the factors affecting the performance of EBT. The overall result indicates the usefulness of the proposed computational translation in biomedical research for human immunity using murine models.