Internal multiples, which mainly contaminate land datasets, are more difficult to address than surfacerelated multiples. Internal multiples generally appear to be very similar to the primaries because of their comparable velocities; they are therefore difficult to differentiate, especially in areas with relatively flat geology. In northern Tarim basin of China, most surveys have this type of flat geology and internal multiple problem. Traditional event discrimination methods, such as multi-channel dip filter, Radon transform, and inside mute, have failed to effectively remove the multiples. We show recent results using a prediction based framework using the extended internal multiple prediction (XIMP) followed by adaptive subtraction, where encouraging results show a clear removal of internal multiples in Northern Tarim basin.