An improved version of the Marine UXO characterization system MBUD, called MBUD2, has been built and tested. The improved version has a new data acquisition system and incorporates technical modifications to eliminate the noise encountered in seawater with the MBUD1 version. The target is illuminated with fields from three orthogonal transmitters, and the signal is measured using four three component receivers. The receivers are arranged in a configuration that inherently cancels the response of the system to enclosing seawater, and to the response of the air-sea interface for shallow deployments. The system was operated as a cued system. It is mounted on a simple platform on top of a support structure that is transported to a predefined survey location by a carrier boat, lowered to the sea bottom, and maneuvered by a diver. The support structure extends one meter below the system platform to allow for the object placement to a specific location by the diver in low-visibility conditions. MBUD2 is able to classify a detected object on the basis of its size, orientation, depth, shape and metal content. The classification is done by estimating the principal dipole polarizabilities of a target using 33 channels of normalized induction responses computed at 46 logarithmically spaced times between 0.0005 and 0.013 s after transient shutoff. Our results clearly show that MBUD2 can resolve the intrinsic polarizabilities of the target, and that there are very clear distinctions between symmetric intact UXO and irregular scrap metal. The system was tested on land and in San Francisco Bay with excellent results. The system meets the design requirements: it cancels the response of the seawater and air-sea interface, the measurements are stable and repeatable, and target responses estimated from marine measurements are identical to those from land measurements.