The invention relates to a random noise suppression method for parallel epicentre seismic data based on PCA-EMD. According to the method, firstly, the spectral range of a useful signal is estimated according to spectrum analysis, secondly, modal components dominated by the useful signal in modal components obtained through EMD decomposition are selected according to the useful signal spectrum range, thirdly, the modal components dominated by the useful signal and the remainder are reconstructed to obtain the reconstruction result, and lastly, a Hankel matrix is constructed for the reconstruction result based on the phase space theory, and PCA decomposition and recovery of the useful signal are further carried out. As verified, the data processing speed is fast, compared with an EMD suppression random noise method, the random noise can be suppressed in the whole frequency range, in the signal-noise band aliasing process, not only can noise energy be suppressed, but also signal details can be effectively protected; through the good signal-to-noise ratio improvement capability, the target data positioning error after processing is smaller, data processing cost is reduced, the qualityof the parallel epicentre seismic data can be effectively improved, and the method is better than other methods under strong voice conditions.