Long reads are reshaping RNA biology. However, determining alternative isoforms from long-read RNA data is a complex and incompletely solved problem even when the reference genome is known. Here we present IsoQuant - a reference-based tool that accurately discovers novel transcripts with at least 3-fold lower false positive rate and 1.8-fold increase in F1-score compared to other tools for Oxford Nanopore data. IsoQuant also increases performance for Pacific Biosciences data.