The influential factors, assessment method, and prediction method for the discrete characteristics of subway vibration source strength were investigated through discrete characteristic analysis. All-day vibration source strength was monitored at 52 typical test sections in subway tunnels under operational conditions in several Chinese cities. The discrete characteristics and main influential factors of the vibration source strength were analyzed. Based on these data, an MLP neural network prediction method was proposed to estimate the subway vibration source strength. The fluctuation amplitude of the measured maximum Z-weighted vibration level exceeds 20 dB. The number of trials employed significantly affects the stability of the assessed vibration source strength. The fluctuation amplitude is 4.5 dB and 2.2 dB for 5 and 20 trains pass-by events, whereas the value reduces to 0.3 dB when 100 trials are adopted. The out-of-roundness of wheels is the main factor that leads to the discreteness of measured vibration levels. Moreover, the proposed MLP neural network prediction method shows high accuracy in predicting the subway vibration source strength.