© 2016 American Chemical Society. Lipoxygenases (LOXs) are a family of enzymes involved in the biosynthesis of several lipid mediators. In the case of human 15-LOX, the 15-LOX-1 and 15-LOX-2 isoforms show slightly different reaction regiospecificity and substrate specificity, indicating that substrate binding and recognition may be different, a fact that could be related to their different biological role. Here, we have used long molecular dynamics simulations, QM(DFT)/MM potential energy and free energy calculations (using the newly developed DHAM method), to investigate the binding mode of the arachidonic acid (AA) substrate into 15-LOX-2 and the rate-limiting hydrogen-abstraction reaction 15-LOX-2 catalyzes. Our results strongly indicate that hydrogen abstraction from C13 in 15-LOX-2 is only consistent with the "tail-first" orientation of AA, with its carboxylate group interacting with Arg429, and that only the pro-S H13 hydrogen will be abstracted (being the pro-R H13 and H10 too far from the acceptor oxygen atom). At the B3LYP/6-31G(d) level the potential and free energy barriers for the pro-S H13 abstraction of AA by 15-LOX-2 are 18.0 and 18.6 kcal/mol, respectively. To analyze the kinetics of the hydrogen abstraction process, we determined a Markov model corresponding to the unbiased simulations along the state-discretized reaction coordinate. The calculated rates based on the second largest eigenvalue of the Markov matrices agree well with experimental measurements, and also provide the means to directly determine the pre-exponential factor for the reaction by comparing with the free energy barrier height. Our calculated pre-exponential factor is close to the value of kBT/h. On the other hand, our results suggest that the spin inversion of the complete system (including the O2 molecule) that is required to happen at some point along the full process to lead to the final hydroperoxide product, is likely to take place during the hydrogen transfer, which is a proton coupled electron transfer. Overall, a different binding mode from the one accepted for 15-LOX-1 is proposed, which provides a molecular basis for 15-LOX-2 exclusive 15-HPETE production in front of the double (although highly 15-) 12/15 regiospecificity of 15-LOX-1. Understanding how these different isoenzymes achieve their regiospecificity is expected to help in specific inhibitor design.