The present work investigates the possibility of constructing a multivariate calibration model for predicting the composition of ground beef with respect to different meat quality types, based on intensity profiles from isoelectric focusing of water-soluble proteins. Beef mixtures containing various amounts of mechanically recovered meat, head meat and production meat from beef, were analysed by isoelectric focusing in immobilised pH-gradients. The gels were photographed and the images transferred to a digital format. By simple image processing procedures, background colour was virtually eliminated and signal strength was improved to a considerable degree. Multivariate analysis of protein profiles from the gels gave models explaining 75 to 90% of variance in sample composition. Manually deboned meat was explained to the highest degree, and with a precision of 7%. Two different qualities of mechanically recovered meat could be detected even when treated as one category. The present approach needs further refinement, but seems applicable for detecting intentional substitution of high quality meat products with low-price raw materials. One advantage of the approach is that evaluation of samples is not dependent on specific knowledge on the individual components to be analysed, so that such analytical methods are relatively easy to implement in any standard laboratory.