NIR spectrometry has become a powerful analytical technique for agro-food products. However, it needs complex transfer to valuate calibration databases, generally based on reference objects, called standards. In the TOP method (Andrew & Fearn, 2004), the alteration of the standard spectra are used to identify the subspace containing most of the perturbations, responsible for the transfer problem. This subspace is then removed from the calibration database, by an orthogonal projection. This paper presents a new method, based on TOP, working without any standard. It only uses the spectra of a few "bridge samples", of known concentration \yc\, that have been measured only on the slave spectrometer. For each yc-bridge sample value, a virtual spectrum is computed from the master database, by applying an yc-centered linear kernel. All the differences between virtual and actual spectra are used to define the subspace to remove by orthogonal projection. The most important advantages of this method are: (i) The orthogonal correction is "embedded" into the model, which means that it is not necessary to preprocess the new spectra when applying the model, (ii) No standard samples are required (iii) The instrumental effect on the spectra can be interpreted through the orthogonal projection loadings. This method has been applied to the transfer from a FOSS NIR6500 database, collected over 10 years on green coffee, towards a FOSS NIR 5000 TR. The studied criteria was: moisture, caffeine and fat content. For each criterion, 4 bridge samples have been used for implementing the correction. Prediction errors, estimated on the slave database, significantly decreased. The biases observed on non corrected predictions totally disappeared. Results were compared with classical standardization protocols. This approach demonstrates the power of orthogonalization methods to transfer and valuate existing calibration databases very easily, without any standard. (Texte intégral)