The pervasive transcription of the human genome results in a heterogeneous mix of coding and long non-coding RNAs (lncRNAs). Only a small fraction of lncRNAs possess demonstrated regulatory functions, making it difficult to distinguish functional lncRNAs from non-functional transcriptional byproducts. This has resulted in numerous competing classifications of human lncRNA that are complicated by a steady increase in the number of annotated lncRNAs. To address these challenges, we quantitatively examined transcription, splicing, degradation, localization and translation for coding and non-coding human genes. Annotated lncRNAs had lower synthesis and higher degradation rates than mRNAs, and we discovered mechanistic differences explaining the slower splicing of lncRNAs. We grouped genes into classes with similar RNA metabolism profiles. These classes contained both mRNAs and lncRNAs to varying degrees; they exhibited distinct relationships between steps of RNA metabolism, evolutionary patterns, and sensitivity to cellular RNA regulatory pathways. Our classification provides a behaviorally-coherent alternative to genomic context-driven annotations of lncRNAs.