The age old paradigm of quantitative structure-activity relationship (QSAR) is the congenericity principle which states that similar structures usually have similar properties.  But these days a lot of large and structurally diverse data sets of chemicals with the same experimental data (dependent variable) are available.  Starting with the same classes of descriptors we extracted the two subsets of the most significant predictors for the formulation of QSARs for two sets of chemicals:  A homogeneous set of 95 amine mutagens and a diverse set of 508 structurally diverse mutagens.  The predictors included calculated topostructural (TS), topochemical   (TC), geometrical, and quantum chemical (QC) indices. Whereas for the homogeneous amines, a small group of descriptors were sufficient for QSAR development, for the 508 diverse set we needed a large and diverse set of indices for effective QSAR formulation.  This empirical study thus vindicates the DIVERSITY BEGETS DIVERSITY paradigm of QSAR.