The added water content of fresh and untreated pork, poultry, fish and prawns was adjusted either by dipping in polyphosphate and salt solutions of known concentrations for controlled periods, or by injection (pork) with polyphosphate and salt solutions. Mixtures were prepared from treated and untreated fish and other foods (milk and flour). Liquid uptake was determined by the weight gain of samples. The proximate composition of the samples was determined: water, fat, protein, NaCl and phosphorus (for polyphosphate content). Complex dielectric spectra of each product were measured at known temperatures and at 31 frequencies in the range 0.2-12 GHz using an automatic network analyser (ANA) and a 3.0 mm open-ended coaxial sensor. The spectra were subjected to various procedures. Principal component analysis (PCA) using the individual complex components. Regression of the composition data against the principal components to obtain prediction formulae for composition including liquid uptake (internal cross-validation). Regression of the composition data against raw spectral data and against other composition variables to obtain similar formulae. In order to design a simpler instrument, the effect on accuracy was studied of reducing the number of frequencies in the spectrum and its range. The slight loss of accuracy engendered by using only five or six frequencies was acceptable. The accuracy of the method in predicting liquid uptake and composition was good. Using one of the compositional variables in the calibration made it equivalent to accuracy obtained by proximate analysis, which was the limiting factor. Measurements on solutions of glycerol, NaCl and water with precisely known composition demonstrated that the intrinsic accuracy of the instrument was far better. A prototype instrument was built and validated using samples of prawns and herring. © 2001 Elsevier Science Ltd.