Robust modeling methods were implemented for the transfer of near-infrared calibration models in intra and inter-brand situations. A network of four instruments from two brands (Foss Infratecs and Bruins OmegAnalyzerGs) was used to implement spectral pretreatment methods, local and variable selection techniques, and orthogonal methods to transfer protein, oil, and linolenic acid models across instruments of the same brand and across instruments of different brands. A total of fifty seven techniques were implemented among which spectral filtering methods based on the smoothing of high frequency components of Fourier and wavelet transforms. A new approach to local similarity was introduced. Results showed that the effectiveness of the various methods was instrument, parameter, and validation set dependent. In some situations, no differences could be observed between master and secondary unit predictions. Local methods appeared to be the weakest methods, most likely due to a problem of over-fitting (specialization) of the calibration set. The transfer of calibrations across brands was possible with performances similar, or better, than in intra-brand calibration transfer.