It is reported that calibration transfer is able to compensate the variations in sensitivity in direct coupling of a headspace sampler to a mass spectrometer when used for quantification purposes using multivariate calibration techniques. This strategy of signal stability compensation allows the use of models constructed from large calibration standard sets without having to repeat their measurement even though variations occur in sensitivity, which may or may not be constant along the mass range. This technique offers advantages over the use of internal standards in this methodology and only requires the measurement of a small number of transfer samples with each set of unknown samples. The results obtained in the determination of six volatile organic compounds-benzene, toluene, ethylbenzene, and m-xylene (BTEX), methyl tert-butyl ether (MTBE), and mesitylene-are reported. To obtain an appropriate calibration set, a Plackett-Burman design with five levels of concentration for each component was employed. A PLS multivariate calibration model was constructed with a group of 25 samples. For selection of the optimum number of principal components, an external validation set (5 samples) was used and the prediction capacity of this set was checked with an additional group of samples that had not been used either in the construction or in the validation of the model. The results obtained can be considered highly satisfactory, and the methodology was successfully tested with natural matrixes (river and tap water).