Fractures are of great interest for hydrocarbon production. In particular, they can provide conduits for fluid flow, hence knowledge of their distribution and orientation can be critical, especially for horizontal wells. Unfortunately, most fractures are relatively small, and well beyond the imaging range of conventional seismic data. On the other hand, it is increasingly recognized that fractures do have a strong influence on shear-wave energy (S waves). Indeed, amplitude variations in stacked (qS2) shear-wave data have successfully been used to identify productive fracture clusters in the Austin Chalk formation in Texas (Mueller 1991). Although seismic information on fracturing is most readily obtained from shear waves, direct measurements of shear-wave energy require both specialized sources and multicomponent receivers, and are not feasible in marine environments, except perhaps at the sea floor. Of importance is the fact that fractures have an impact on the entire seismic wavefield, with the imprint of the fracture symmetries leading to a general azimuthal behaviour. As a result, although compressional (P) waves are relatively insensitive to fracturing at near offsets, they can become linked to the larger effects experienced by the shear-waves through mode conversion at oblique incidence, and also through their own specific sensitivity. Thus, it is feasible that anisotropic variations in P-wave behaviour can provide a way of obtaining information on fracture properties at subseismic scales. (MacBeth 1995). This then offers a potential means of studying fracture distributions and orientations in marine areas. In this paper we present evidence to show that information on both the presence and nature of fractures can be obtained from conventional marine (i>P-wave) seismic data using AVO (amplitude variation with offset) analysis. In particular, our results demonstrate that fracture-related AVO has a distinct anisotropic (directional) behaviour that can be linked directly to the azimuthal orientation of the fractures. Indeed, this is consistent with the numerical study of Mallick & Frazer (1991), and has previously been observed in land data for a variety of fractured gas reservoirs (Lynn et al. 1996). However, to our knowledge, the results given in this paper constitute the first example of this phenomenon for a marine dataset.