Near Infrared instruments today are conveniently used by plant breeders and industry as spectral 'multimeters' for instant analysis of a range of chemical components. These are based on multivariate computer calibrations by chemometrics (e.g. Partial Least Squares Regression - PLSR - and neural nets) facilitated by the instrument suppliers. Here we will demonstrate in barley how the near infrared spectra that are routinely measured for specific chemical screening by breeders can now be exploited for chemical-physical characterisation of whole samples or genotypes. From near infrared data in a Principal Component Analysis (PCA) score plot the breeder can now directly select cereal seed genotypes with improved chemical-physical quality by 'data breeding'. This is achieved by selecting recombinants which are positioned close to high quality control genotypes in the score plot. Differences in local spectra can be deduced by PLSR to physical/chemical components by destructive analyses. Mathematical data compression models are useful. However, they are all destructive to the finely tuned, highly reproducible spectral patterns that originate from the self-organised biological network represented by the developing barley endosperm. The evaluation of Near Infrared Reflectance Spectroscopy (NIRS) data must therefore include a final direct visual inspection of spectra to trace reproducible patterns of biological significance from individual phenotypes. The 'LatentiX' graphic data interface that is demonstrated at the symposium is constructed to visualise spectral areas correlated to specific chemical components and genotypes. It is available at www.latentix.com