The relative precisions of five multivariate calibration methods [direct multicomponent analysis (DMA), stepwise multiple linear regression (SMLR), principal components regression (PCR), and PLS1 and PLS2 (PLS = partial least-squares)] are evaluated for the determination of transitionmetal ions in model multicomponent systems. These systems represent simulated industrial process streams containing mixtures of two, three and five metal ions. Physical and chemical interferences have been incorporated to provide a rigorous test of the calibration techniques. Diode-array spectrophotometry has been used to obtain spectra of the inherent absorbances of the metal ions in the visible region. Multivariate calibration models have been constructed from these data and used to predict concentrations of metal ions in test solutions. Results are presented for ‘well behaved’ and more complex multicomponent systems, and the predictive performance of each calibration technique is discussed. It is demonstrated that SMLR, PCR, PLS1 and PLS2 provided calibration models significantly more robust than those of DMA when physical and chemical interferences were present. SMLR often provided the best precisions in both well behaved and less well behaved systems. Calibrations based on first-derivative data afforded greater precision than those based on absorbance data.