Traffic analysis is considered the most powerful strategy of disclosing the hidden communication relations in an anonymous communication system. Statistical traffic analysis attacks are even more subtle in that the attackers are usually eavesdroppers who do not modify the network's behaviors. Moreover, the attackers even do not need to look into the traffic content, which may be encrypted, in order to analyze the statistical characteristics. Such attacks have been thoroughly investigated for static wireline networks. However, none of these mechanisms can be directly applied to mobile ad hoc networks (MANETs) due to the inability to deal with mobility, the ad hoc infrastructure and the broadcasting nature of wireless transmissions. Recent research conducted on statistical traffic analysis attacks targeting MANETs is restricted to disclosing the end-to-end traffic distribution. In this paper, we present the least squares disclosure attack (LSDA), targeting a popular MANET routing strategy, that is, the position based routing (PBR, a.k.a geographic routing). LSDA utilizes the traffic distribution disclosed by existing solutions, and de-anonymizes the network communication on a per-flow basis by identifying the source and destination of each end-to-end flow. In LSDA, traffic disclosure is modeled as an efficiently solvable least squares problem subject to linear constraints. The empirical study demonstrates that, the proposed solution can de-anonymize the network flows in high accuracy.