As the land internal multiples could not been obviously attenuated, the authors develop the dimensionality reduced virtual event method to suppress the internal multiples in pre-stack land seismic data. Compared with the traditional virtual event method, the authors apply the virtual event method trace by trace on the pre-stack gathers after accurate dynamic correction to predict internal multiples. It achieves dimensionality reduction, greatly reduces the amount of calculation, and no longer requires a regular and dense enough distribution. Meanwhile, this method introduces a weighted reference trace with high signal-to-noise ratio to participate in the cross-correlation and convolution operations of the pre-stack gathers, which improves the prediction accuracy of the virtual event method. This method is applied to actual land seismic data in western China and achieved obvious effect when suppressing internal multiples.