In the summers of 2009 and 2010, simultaneous multi-station observations of natural lightning were conducted in the Da Hinggan Ling forest region of China using a network primarily consisting of fast and slow antennas. In 2010, 37 positive and 56 negative cloud-to-ground (CG) lightning flashes from three thunderstorms were selected and their corresponding preliminary breakdown pulse (PBP) trains were statistically analyzed simultaneously. Several parameters were determined: the total duration of the PBP train (Total Duration), the separation between the PBP train and the first return stroke (PB-RS Separation), the interval between the PBP train and the first return stroke (Pre-RS Interval), the individual pulse duration of the PBP train (Individual Pulse Duration), and the interval between successive pulses of the PBP train (Interpulse Interval). For negative CG lightning, the arithmetic means (AMs) of these parameters were 4.1 ms, 55.4 ms, 56.0 ms, 8.8 μs, and 111.0 μs, respectively, and the geometric means (GMs) were 3.7 ms, 35.6 ms, 36.5 ms, 7.4 μs, and 98.2 μs, respectively. For positive CG lightning, the respective AMs were 4.5 ms, 75.6 ms, 77.3 ms, 11.5 μs, and 297.3 μs, and the respective GMs were 3.0 ms, 57.8 ms, 60.0 ms, 10.0 μs, and 217.9 μs. PBP trains in positive CG lightning were found to exhibit longer durations and longer intervals preceding the RS (Return Stroke) than those for PBP trains in negative CG lightning. Moreover, single pulses of PBP trains in positive CG lightning were generally wider and arranged more sparsely throughout the sequences than their negative counterparts. The ratios between the maximum PBP and the RS (PB/RS, Preliminary Breakdown) for both negative and positive CG lightning were calculated. Comparison of these ratios with the results of previous studies suggests that the ratios for negative CG lightning increase with latitude, whereas those for positive CG lightning do not. In addition, the samples were investigated to ascertain whether they were in agreement with the so-called BIL model, and the results obtained here were compared with those of other studies.