If rollovers during play are acts of submission, then we would expect them to be (i) triggered by behavior that is overtly aggressive or that causes detectable discomfort or pain, (ii) performed predominantly by one of the play partners, who will be disadvantaged by being smaller or weaker, and (iii) the supine position will be sustained, thereby (iv) inhibiting the play behavior or aggression of the other animal. If, on the other hand, they are executed tactically, for combat purposes, then they should (i) be triggered by an attack and (ii) serve to block it or (iii) lead to an immediate attack or counter-attack. They are therefore either likely (iv) to be performed by either partner or (v) by the larger animal in the context of soliciting play. They will therefore (v) not inhibit the continuation of play. Finally, if used for playful solicitation, rather than inhibiting attack, rollovers should elicit playful attack by the nearby partner. To test these predictions, we analyzed two sets of videotaped footage of the play bouts of adult dogs.