Although social animals frequently make decisions about when or with whom to cooperate, little is known about the underlying mechanisms of partner choice. Most previous studies compared different dyads' performances, though did not allow an actual choice among partners. We tested eleven ravens, Corvus corax, in triads, giving them first the choice to cooperate with either a highly familiar or a rather unfamiliar partner and, second, with either a friend or a non-friend using a cooperative string-pulling task. In either test, the ravens had a second choice and could cooperate with the other partner, given that this one had not pulled the string in the meantime. We show that during the experiments, these partner ravens indeed learn to wait and inhibit pulling, respectively. Moreover, the results of these two experiments show that ravens' preferences for a specific cooperation partner are not based on familiarity. In contrast, the ravens did show a preference based on relationship quality, as they did choose to cooperate significantly more with friends than with non-friends and they were also more proficient when cooperating with a friend. In order to further identify the proximate mechanism of this preference, we designed an open-choice experiment for the whole group where all birds were free to cooperate on two separate apparatuses. This set-up allowed us to distinguish between preferences for close proximity and preferences to cooperate. The results revealed that friends preferred staying close to each other, but did not necessarily cooperate with one another, suggesting that tolerance of proximity and not relationship quality as a whole may be the driving force behind partner choice in raven cooperation. Consequently, we stress the importance of experiments that allow such titrations and, suggest that these results have important implications for the interpretations of cooperation studies that did not include open partner choice.