Long non-coding RNAs (lncRNAs) have recently emerged as key regulators of gene expression in embryos and in embryonic stem cells (ESCs). Recent large-scale genomics approaches have identified thousands of putative lncRNAs but are these all truly non-coding RNAs? Here, on p. 2828, Alex Schier, Eivind Valen and colleagues set out to answer this question. The researchers use ribosome profiling to identify translated transcripts, combined with a machine-learning approach to classify open reading frames (ORFs) and to validate zebrafish lncRNAs. They find that many proposed lncRNAs are in fact protein-coding contaminants. Moreover, their study reveals that many zebrafish and ESC lncRNAs resemble the 5’ leaders of coding RNAs, suggesting a novel mechanism for lncRNA regulation. Overall, the findings presented here clarify the annotation of lncRNAs, as well as offering a valuable resource that can be used for identifying translated ORFs and hence novel protein-coding genes that function during zebrafish development.