Picking up arrival time of P-wave is the key to study on seismic source location and seismic tomography.Aiming at the shortcomings of redundant information

noise in seismic data and the accuracy of picking up P-wave is not high

P-wave automatic picking method is studied. By improving local linear embedding（LLE）algorithm

dimension reduction processing of data samples are carried out. Through studying on different measurement method and the inner product calculation method of time series

a method of kernel classification for time series is proposed.Combining the above methods

an automatic method for picking up arrival time of P-wave（AMPAT） based on multidimensional feature is proposed.15-dimensional feature attributes are extracted from seismic signal as data samples.And the LLE method is utilized to extract the data samples in order to reduce the dimension of the data.P wave arrival time is determined by the nuclear classification method.The simulation results show that accuracy of the AMPAT is improved obviously compared with short term average/long term average（STA/LTA） method and maximum likelihood（ML） method.