Teleseismic location is heavily dependend on the spatial distribution of recording stations and the number of registered phases. In the Arctic Basin, a total lack of local or regional recording stations makes location of earthquakes especially challenging.In 1999 an unusually strong earthquake swarm at Gakkel Ridge, Arctic Ocean, which consisted of 252 events with body-wave magnitudes up to 5.2 lasted over 9 months. Using the ISC reviewed dataset, we relocated this swarm with the probabilistic routine NonLinLoc and tested extensively the influence of velocity model, station coverage and weighting on the location result. For calculation of travel times we used the velocity model AK135 combined with a regional velocity model for recording stations with epicentral distance < 30°. T he dataset was reviewed with regard to the quality of the location and reduced to 63 well located events whose epicentre locations are largely independent of the above tested parameters. T he 68% error in ellipse semi-major axes of the new localization is in the order of 15 km for the best events.We tested if the choice of location routine influences the resulting locations. Comparing our solution for the 1999 earthquake swarm to that for the L2-norm-based algorithm Hyposat, we found significant differences in the location for the 63 best constrained events, the epicenters not even matching within error-ellipses. Additionally, to improve locations within the cluster, we performed a simple master event relocalization and tested again the dependence of the results on the location algorithms.