A fundamental need for modern biomechanical and clinical application is the ability to accurately capture normal and pathologic human movement. In this contest, nowadays, the aquatic environment is gaining an even important role within the rehabilitation world, where one of the most common motor task is walking in water. However, state of the art of marker based gait analysis cannot be performed in water environment, where the attachment of skin surface markers is not easy. A markerless approach, based only on synchronized video sequences, has been proposed as a potential solution for underwater (UW) applications. In the present work, an automatic markerless motion capture system has been investigated and its accuracy in UW three-dimensional (3D) lower limbs joint kinematics reconstruction has been tested by means of a comparison with the state of the art. The results demonstrate the feasibility of calculating meaningful joint kinematics from subjects walking UW without any markers attached to the limb. Thus, a markerless approach seems to offer the promise of expanding the applicability of human motion capture in an aquatic environment.