This paper is an endeavour to address the problem of distributed leader selection in a formation of autonomous agents where the agents do not communicate directly via communication channels. The algorithm that the agents use to select a leader relies on the agents observing each others' behaviours. It is shown that the proposed algorithm is terminated, on average, in finite number of step and results in the selection of a leader for the formation. Moreover, It is established that the algorithm has some common elements with an algorithm widely used in data networks, i.e.~Slotted Aloha. The application of the algorithm to a formation controlled by a nonlinear control law is studied and some numerical examples are presented to show the general performance of the algorithm.