Nearest neighbor distributions of molecular spectra can, in principle, be used to learn from quantum spectra about the classical dynamics of a system, i.e., whether it is regular or irregular (chaotic). However, the predictive power of this method is limited due to the generally small number of spectral lines available for analysis, and the ambiguities of the procedures used. This is demonstrated here for the determination of the shape of nearest neighbor distributions in terms of a Brody parameter, which was determined from fits to samples from a Brody distribution and fits to simulated molecular spectra. The procedures are also applied to computed spectra of NO2 and SO2.