**A harbour seal (Phoca vitulina) can learn geometrical relations between landmarks** Data for the paper at JEB Authors: Eric Maaß, Lars Miersch, Gerit Pfuhl, Frederike Hanke The files contain videos, training-data, testing-data and the statistical analysis in JASP (see jasp-stats.org) The present study aimed to evaluate, whether harbour seals could find a goal using relational information from two landmarks in form of a relational rule, a “middle” rule, i.e. the goal was always in the middle of the landmarks irrespective of their orientation or the inter-landmark distance. This study built upon a previous experiment, in which we could show that seals spontaneously encode goals in respect to landmarks as vectors or use landmarks as beacons (Maaß & Hanke 2022). For this experiment, we used the sophisticated setup, a large underwater LED panel we had designed for testing goal localization in respect to landmarks in our seawater enclosure for the first landmark study (Maaß & Hanke 2022). During the current study, this setup unfortunately irreparably collapsed on July 9th, 2021. At this point, we had completed data collection for seal “Moe” as presented in this manuscript; training had to be stopped for two additional seals in early stages of their training. The results of seal “Moe” clearly show that this seal was able to learn to find a goal in the “middle” of two landmarks when presented with novel (fully or partially novel) two LED landmark arrays. Adopting a relational middle rule could especially be beneficial when using landmarks from afar.