Histone modifications play important roles in regulating eukaryotic gene expression and have been used to model expression levels.Here, we present a regression model to systematically infer mRNA stability by comparing transcriptome profiles with ChIP-seq of H3K4me3, H3K27me3 and H3K36me3.The results from multiple human and mouse cell lines show that the inferred unstable mRNAs have significantly longer 3Untranslated Regions (UTRs) and more microRNA binding sites within 3UTR than the inferred stable mRNAs.Regression residuals derived from RNA-seq, but not from GRO-seq, are highly correlated with the half-lives measured by pulse-labeling experiments, supporting the rationale of our inference.Whereas, the functions enriched in the inferred stable and unstable mRNAs are consistent with those from pulse-labeling experiments, we found the unstable mRNAs have higher cell-type specificity under functional constraint.We conclude that the systematical use of histone modifications can differentiate non-expressed mRNAs from unstable mRNAs.In summary, we represent the first computational model of mRNA stability inference that compares transcriptome and epigenome profiles, and provides an alternative strategy for directing experimental measurements.Background: Histone modifications are predictive of gene expression in multiple cell lines, so it may reflect steady-state transcription rate.However, the expression index, measured by either microarray or RNA-seq, represents the balance between transcript production and degradation.Our study is aimed to illuminate the biological meaning of the difference between mRNA level and histone modifications level.Methods: Using the information from histone modifications-H3K4me3, H3K27me3 and H3K36me3, we derive regression model to fit mRNA level in multiple cell lines.The mRNAs, whose expression levels (true values) are significantly lower than histone modifications level (prediction value), are inferred as unstable mRNAs (faster degraded), and vice versa.Results: Inferred unstable mRNAs harbor sequence signatures associated with more microRNA targeting.Half-lives are highly correlated with residuals in an RNA-seq model but independent of those in a GRO-seq model.Inferred unstable and stable mRNAs are enriched for distinct biological processes, consistent with previous study.Inferred unstable mRNAs have higher cell-type specificity under functional constraint Conclusions: The difference between histone modification levels and mRNA levels implies mRNA stability.The lowly expressed mRNA are actually divided into two group, the majorities are silenced mRNAs, and the remainings are unstable ones .