26) Shaka, A. J.; Keeler, J.; Frenklel, T.; Freeman, R. J. Magn. Res. (27) Guiilou, C.; Trlerweiler, M.; Martin, G. J. Mgn. Reson. Chem. 1988, (28) Martln, 0. J.; Naulet, N. Fresenlus' 2. Anal. Chem. 1988. 332, (29) Craig, H. Geochlm. Cos". Acta 1957, 12, 133-149. (30) Rauschenbach, P.; Simon, H.; Stichler, W.; Moser, H. Z. Mturforsch. (31) Misselhorn, K.; Bruckner, H.; MussigZuflka, M.; Grafahrend, W. 1990, 94. 8303-8309.ThkmMlecrlptdescdbwmethodsfor detecting and modeling nonlinear regions of spectral response In multivariate, multicomponent spectroscopic assays. Slmuiated data and ex-parimenta1 UV/vklble data were used to 8tudy the capablilty of multivariate Ilnear models to approximate nonlinear response. Tho sources of real and apparent noniinearlty dmulated Included nonilnear Instrument response functions (e.g. dray light), concentrationdependent wavelength shifts, and concentratlon dependent absorption bandwidth changes. A weighting algorithm was devised to reduce the Influence of nonlnear spectral reglons in principal component regresdon (PCR) caiibratlons, thereby improvlng the performance of muitivariate linear calibration models. Secondorder caiibrat h methods using quadratic principal component scores and nonilnear calibration methods using artlficlal neural networks were compared to unwelghted and welghted i h a r callbration methods. Orthogonal transformation of the input variables was used to dgnlficantly improve neural network trainlng speed and reduce callbration error. Some conditions where wc-r and nonitnear calibration techniques outperform Ilnear callbration techniques have been ldentlfled and are dercribod.