Lasso peptides are a class of knot-like polypeptides in which the C-terminal tail of the peptide threads through a ring formed by an isopeptide bond between the N-terminal amine group and a side chain carboxylic acid. The small size (∼20 amino acids) and simple topology of lasso peptides make them a good model system for studying the unthreading of entangled polypeptides, both with experiments and atomistic simulation. Here, we present an in-depth study of the thermal unthreading behavior of two lasso peptides astexin-2 and astexin-3. Quantitative kinetics and energetics of the unthreading process were determined for variants of these peptides using a series of chromatography and mass spectrometry experiments and biased molecular dynamics (MD) simulations. In addition, we show that the Tyr15Phe variant of astexin-3 unthreads via an unprecedented "tail pulling" mechanism. MD simulations on a model ring-thread system coupled with machine learning approaches also led to the discovery of physicochemical descriptors most important for peptide unthreading.