RNA-sequencing is a powerful approach providing estimates of both isoform and gene expression with unprecedented dynamic range and accuracy. A fundamental goal of RNA-seq experiments measuring expression in two or more biological conditions is the identification of differentially expressed isoforms and genes. Most of the statistical methods developed to identify differentially expressed genes measured using microarrays do not directly apply, and the methods that have been developed specifically for RNA-seq measurements do not directly accommodate isoform level expression, dependence across isoforms, and mapping uncertainty. We have developed an empirical Bayesian modeling approach that accounts for and capitalizes on these features. Motivation for and advantages of the approach are illustrated in simulations, in RNA-seq studies of human embryonic stem cells, and in an RNA-seq study of mammary carcinogenesis in a rat model for breast cancer.