RNA sequencing (RNA-seq) is a powerful approach for measuring gene expression levels in cells and tissues, but it relies on highquality RNA. We demonstrate here that statistical adjustment using existing quality measures largely fails to remove the effects of RNA degradation when RNA quality associates with the outcome of interest. Using RNA-seq data from molecular degradation experiments of human primary tissues, we introduce a methodquality surrogate variable analysis (qSVA)-as a framework for estimating and removing the confounding effect of RNA quality in differential expression analysis. We show that this approach results in greatly improved replication rates (>3×) across two large independent postmortem human brain studies of schizophrenia and also removes potential RNA quality biases in earlier published work that compared expression levels of different brain regions and other diagnostic groups. Our approach can therefore improve the interpretation of differential expression analysis of transcriptomic data from human tissue.RNA sequencing | differential expression analysis | statistical modeling | RNA quality M icroarrays and RNA sequencing (RNA-seq) can measure gene expression levels across hundreds of samples in a single experiment. As gene expression levels are measured with error, normalization procedures have been implemented for both microarray (1) and RNA sequencing (2) data to reduce technical variability, including controlling for variability associated with how and when the samples are run, so-called "batch" effects (3). Recent research has further characterized this expression variability in RNA-seq data (4-6), including demonstrating variability associated with technical factors involved in the preparation, sequencing, and analysis of samples. Variability in gene expression is particularly influenced by RNA quality (7) because accurately measuring gene expression levels strongly depends on the quality of the input RNA. This suggests that a portion of traditionally measured latent "batch" effects could actually be attributed to the underlying quality of the input RNA.Postmortem studies typically extract RNA from tissue that has been susceptible to a wide variety of antemortem and postmortem factors. Several approaches exist for quantifying the quality of the input RNA before sequencing library construction, including UV absorption ratios of 280 nm to 260 nm and RNA integrity numbers (RINs). RIN is a machine learning-derived measurement resulting from placing RNA on a Bioanalyzer and obtaining a tracing of fragment sizes per sample. RIN ranges from 10 (very high quality RNA) to 0 (completely degraded RNA), and the apparent intactness of ribosomal RNAs (which are two large peaks in the fragment size tracing) is one of the most discriminating factors that distinguishes very high quality from moderate quality RNA (8). Recommended RIN thresholds for sample exclusion before data generation have been suggested as low as 5.0 for PCR (7) and 7.0 for RNA-seq. (9). However, even high quality samples (RIN ...