A multivariate method for predicting state of charge, from electrochemical data, of a nickel-metal hydride (NiMH)-battery is presented. Partial least square (PLS) regression is used to evaluate electrochemical impedance spectra and predict state of charge. The impedance spectra are analysed in the frequency range 239-0.6 Hz. The impedance is measured for different states of charge at open-circuit conditions and during continuous discharge at loads ranging between 0.2 C and 0.8 C. When measuring the impedance during discharge, the AC-current signal is imposed on the DC-current. The predictive capability of the method is tested by a cross validation procedure and the root mean square error of prediction is 7% when using the outlier identification capability of the PLS-regression method. The state of charge is evaluated with a single model, independently of whether the cell is subjected to open-circuit or polarised conditions. The predictive performance of the present model decreases at state of charge values less than 10%.