In this paper we present a generalisation of previously considered Markovian models for Tennis that overcome the assumption that the points played are i.i.d and includes the time into the model. Firstly we postulate that in any game there are two different situations: the first 6 points and the, possible, additional points after the first deuce, with different winning probabilities. Then we assume that the duration of any point is distributed with an exponential random time. We are able to compute the law of the (random) duration of a game in this more general setting.
