U.S.A., Multiples, diffractions and their multiples are a common feature of marine seismic data. In some areas of the South China Sea, the residual multiples are a significant problem as they are coincident with the reservoir section. In such cases, we need to devote additional resources to further attenuate the multiples, particularly the diffracted multiples. The challenge is to assess how much effort is sufficient and whether that effort is required during data acquisition, data processing, or both?The work presented in this paper is the result of a real problem that we faced a couple of years ago when planning a seismic survey in the South China Sea. In the absence of 3D seismic data from nearby, the challenge was to demonstrate that 2D SRME would be inadequate in this area and that denser data acquisition would be required to attenuate multiples sufficiently to allow the 3D seismic data to be used for development drilling.The requirement for 3D SRME in areas with rugose seabeds or other multiple generating interfaces is well-documented (van Borselen et al 2005, McHugo et al 2009 and Hung et al 2010 for example).That is one aspect of the challenge faced in the area of this case study: the seabed is locally steeply-dipping and there are shallow diffractors that generate structurally complicated multiples. Both of these factors indicate that 3D SRME would be more effective at attenuating multiples on data from this area than 2D SRME; this requirement was successfully argued using analogue data examples.It is well understood that aliased multiples cannot be effectively attenuated during data processing. We used 2D lines where the seabed multiple was well-attenuated and only the diffracted multiples were troublesome to conduct a decimation test. Ambient noise was attenuated then traces were dropped without an anti-alias filter from 6.25m to give a trace spacing of 12.5m, 18.75m and 25m in the shot domain. This simulates the cross-line trace interval that would be achieved with 50m, 75m and 100m streamer spacing and dual source flip-flop acquisition geometry. Using this technique we were able to demonstrate that the minimum acceptable streamer spacing would be 75m, assuming that one pass of trace interpolation could be used prior to demultiple processing.