We investigated whether the growls of domestic dogs, Canis familiaris, exhibit predictable variations in a range of key acoustic parameters when they are given in two contrasting experimental contexts, and whether humans are sensitive to such variation. In a standardized experimental paradigm, an experimenter visited 32 domestic dogs and generated an aggression context and a play context. In these contexts, 204 isolated growls were recorded and subsequently acoustically analysed. Contrary to previous findings on barks, fundamental and formant frequencies of growls did not vary between the two contexts. However, growls from the aggression context were significantly longer than growls from the play context. Additionally, the temporal structure of the vocal sequences containing growls differed significantly between the contexts. In a series of psychoacoustic experiments, human listeners were not able to discriminate between growls recorded in the two contexts, however both aggressiveness and playfulness ratings were strongly influenced by dog weight, with larger dogs being perceived as more aggressive than smaller dogs. However, participants were able to attribute aggressiveness and playfulness to synthesized sequences in which growls occurred at a two rates, typical of either the aggression or the play context. We conclude that context-related variation in growling behaviour resides in the temporal structure, rather than in the acoustic composition of isolated growls, and furthermore that human listeners are able to attribute context on the basis of growling rate. In addition, human appear to have an intuitive acoustic bias toward perceiving larger dogs as more aggressive.