Background: MicroRNA (miRNA) expression can be characterized by small RNA sequencing. Typically, the resulting sequencing reads are aligned to known miRNAs, and read counts of sequences that align to the same miRNA are summed to produce miRNA-level read counts. The aggregated miRNA counts are then used for differential expression (DE) analyses, often using tools developed for mRNA-seq data analysis. There are two key drawbacks to this approach: (1) aggregation to miRNA-level counts discards information about different miRNA transcript isoforms, called isomiRs, and (2) miRNA-seq data violate key assumptions of the DE methods developed for mRNA-seq data. Together these necessitate the development of DE methods designed specifically for miRNA-seq data that can model isomiR-level data. Methods: We establish miRglmm, a DE method, implemented in an R package of the same name, that uses a generalized linear mixed model (GLMM) of isomiR-level counts to account for dependencies due to reads coming from the same sample and/or from the same isomiR. The isomiR random effect variances can be used to quantify variability in differential expression between isomiRs that align to the same miRNA, thereby facilitating detection of miRNAs with differential expression or differential isomiR usage. Results: Using biological data to simulate miRNA with variable isomiR DE, we demonstrate that miRglmm provides markedly lower mean squared error (MSE) and much better confidence interval coverage than other DE tools. Using an experimental benchmark data set of synthetic miRNAs, we show that even in the absence of isomiR variability, miRglmm provides the lowest MSE among all DE tools while maintaining confidence interval coverage at the nominal level. In real biological data, miRglmm finds significant isomiR-level DE variability for most miRNAs in our analyses. While fold change estimates are similar to those from commonly used DE tools (intraclass correlation coefficients ≥ 0.8), miRglmm differs the most from existing methods and is able to estimate both isomiR- and miRNA-level DE. Conclusions: The use of isomiR-level counts, rather than summed miRNA-level counts, in a GLMM framework, represents a notable improvement in miRNA differential expression analysis. Our method, miRglmm, outperforms current DE methods in estimating DE for miRNA, whether or not there is significant isomiR variability, and simultaneously estimates isomiR-level DE.