The performance of Transmission Control Protocol (TCP) over Ad hoc wireless networks (or simply ad hoc networks) has been extensively studied through simulations by the research community. Although many theoretical models, such as [1], have been proposed for estimating the performance of TCP over wired networks, researchers have faced many difficulties in modeling TCP over ad hoc networks. These difficulties are mainly due to the behavior of the underlying physical and MAC layers. Recently, [2] attempted to solve this problem by simplifying the behavior of TCP, besides assuming that no packet losses occur. In this work, we attempt to provide a theoretical model for TCP by considering the main phases of TCP, namely the slow start phase and the congestion avoidance phase, thus providing a more accurate model that captures all of its main features. To the best of our knowledge, ours is the first model that considers the slow start phase while analyzing TCP's performance in ad hoc networks. We make use of multi-dimensional Markovian chains to model each of these phases. We then use the resulting steady state probabilities to estimate the goodput. Furthermore, the analysis is validated by comparing the theoretical and simulation results using various error models.