As the next step it is necessary to validate multivariate models on the external prediction set of authentic samples and their blends, which were not used for calibration of regression models. To do this the external validation data set I was used, which consisted of heparin blends with 1.9–11.4% w/w of bovine content. In these mixtures low bovine content was deliberately spiked in porcine heparin to test the multivariate models. RMSEP ranged from 0.9% for PLS to 2.6% to SR demonstrating that constructed models are applicable for new heparin samples (Table 1). Good fitness was obtained for PLS (R2 = 0.91), RR (R2 = 0.96), and SPCR (R2 = 0.86) models, whereas SR failed to provide reliable quantitative results (R2 = 0.62). As an example, Fig. 5 showed the predicted-reference plot along with 95% prediction bands for bovine content in blends obtained by PLS regression. The values obtained by leave-out-one cross validation are shown for comparison on the same plot. Though prediction bands for external independent set were broader than those for leave-out-one cross validation, new blends were reliably recognized and quantified by PLS method.