While animals navigating the real world face a barrage of complex sensory input, their brains have evolved to perceptually compress multidimensional information by selectively extracting the features relevant for survival. For instance, communication signals supporting social interactions in several mammalian species consist of acoustically complex sequences of vocalizations, however little is known about what information listeners extract from such timevarying sensory streams. Here, we utilize female mice's natural behavioural response to male courtship songs to evaluate the relevant acoustic dimensions used in their social decisions. We found that females were highly sensitive to disruptions of song temporal regularity, and preferentially approached playbacks of intact male songs over rhythmically irregular versions of the songs. In contrast, female behaviour was invariant to manipulations affecting the songs' sequential organization, or the spectrotemporal structure of individual syllables. The results reveal temporal regularity as a key acoustic cue extracted by mammalian listeners from complex vocal sequences during goal-directed social behaviour.