A performance-level seismic motion hazard analysis method includes: (1) extracting seismic motion data and denoising the data; (2) extracting feature parameters from the data, and carrying out initialization; (3) generating a training set, an interval set and a test set; (4) training a multi-layer neural network based on the training set; (5) training output values of the neural network based on the interval set, and calculating a mean and a standard deviation of relative errors of the output values; (6) training the neural network based on the test set to determine output values, and calculating a magnitude interval based on an interval confidence; (7) carrying out probabilistic seismic hazard analysis to determine an annual exceeding probability and a return period of a performance-level seismic motion; and (8) determining a magnitude and an epicentral distance that reach the performance-level seismic motion based on the performance-level seismic motion and consistent probability.