A subjective evaluation method for the air-borne sound insulation of vehicle body in reverberation room is developed and the correlation between the subjective preference and objective noise reduction level (NRL) is investigated in this paper. The stationary vehicle's interior noise is recorded by using a digital artificial head under a given white noise excitation in the reverberation room, which demonstrates more credible than those in traditional road test methods. The recorded noises of six different vehicles are replayed and evaluated subjectively by 22 appraisers in a sound quality room. The paired comparison scoring method is employed and the check and statistic methods for the subjective scores are introduced. The subjective preference is introduced and calculated by the statistics and normalization of the effective scores, which can indicate an overall preference ranking of all the six vehicles numerically. Furthermore, an objective prediction model is established based on the correlation analyses and linear regressions. The subjective preference is proved to be attributed to the average NRL in 2k-5kHz frequency range only. The subjective evaluation method and the prediction model provide the guidance for the evaluation, prediction, target setting and optimization of the vehicle sound insulation.