RNA-seq and small RNA-seq are powerful, quantitative tools to study gene regulation and function. Common high-throughput sequencing methods rely on polymerase chain reaction (PCR) to expand the starting material, but not every molecule amplifies equally, causing some to be overrepresented. Unique molecular identifiers (UMIs) can be used to distinguish undesirable PCR duplicates derived from a single molecule and identical but biologically meaningful reads from different molecules. We have incorporated UMIs into RNA-seq and small RNA-seq protocols and developed tools to analyze the resulting data. Our UMIs contain stretches of random nucleotides whose lengths sufficiently capture diverse molecule species in both RNA-seq and small RNA-seq libraries generated from mouse testis. Our approach yields high-quality data while allowing unique tagging of all molecules in high-depth libraries. Using simulated and real datasets, we demonstrate that our methods increase the reproducibility of RNA-seq and small RNA-seq data. Notably, we find that the amount of starting material and sequencing depth, but not the number of PCR cycles, determine PCR duplicate frequency. Finally, we show that computational removal of PCR duplicates based only on their mapping coordinates introduces substantial bias into data analysis.