Well that's kind of a big topic....
Broadly your main approaches are:
- treat a (short period) multiple as a reverberating, ringy wavelet; this is where deconvolution in the XT or Linear Tau-P domains works
- model the multiple are remove it with adaptive subtraction; these are methods like "free surface" demultiple
- remove the multiple on gathers based on its apparent velocity, using things like the FK or Parabolic radon domain
- flatten the multiple (on a stacked section or offset planes) and remove the flat energy
These all work well on marine mutliples, and tend to be more complex on land ones.
Typically on a marine dataset you would use SRME (free surface) demultiple to hit longer periods (>200ms), then tau-P domain deconvolution for shorter periods, followed by a residual parabolic radon multiple for any pesky multiple tails the SRME had missed.
Of course, a lot depends on your specific problem; diffracted multiples tend to be hard, and the issue is often that there is not very much signal under the multiple once you have removed it.
There's also a lot of theme-and-variation, so for example you can do Radon demulitple by applying a 90% of primary NMO correction to gathers (primaries are up dip, multiples downdip) but in some situations you might want to apply a water-velocity NMO or even pick a mulltiple trend and remove things that way.
There's a description of Radon here : http://seismicreflections.globeclaritas ... iples.html
If you can post some data I can make some suggestions...