This research investigates the circadian rhythms of friend-making behavior using large-scale data from a social media platform, in particular a popular online gaming community. Enlightened by the notion of human biological clock and afforded by the massive yet fine-grained data, we longitudinally track the daily changes in actual friend-making activities of participants in the community and uncover regularities in this pertinent social behavior. We show that people are most likely to make friends at the night (e.g., 20:00, 0:00) and the least likely to do so in the morning (e.g., 8:00). This pattern was consistently observed after considering the number of available players, the players’ game levels, the effect of weekend, and time zones. The systematic variation unveiled in people’ friend-making behavior by hour of a day deepens our scientific understanding of this integral social behavior of human everyday life.