ABSTRACT An empiric procedure of successive approximations for the selection of the most convenient wavelengths and instrumental parameters to execute an UV multivariate analysis is described. The selection criterion was according to the minimum PRESS value. For the identification of the best wavelength intervals a tree-fraction scheme has been used, starting from the full spectrum and going down to 3 nm intervals. General considerations for applying this procedure on any multicomponent mixture are finally reported. The study has been developed on a mixture of propyphenazone, paracetamol and caffeine, present in several pharmaceuticals, whose spectra so totally overlap to make inconsiderable the classical spectrophotometric methods. The obtained results were enthusiastic demonstrating a very high improvement of accuracy and precision. In fact the mean error on propyphenazone, paracetamol, caffeine passed respectively from 12.8, 5.49, 41.88 %, reading the full spectrum at the default conditions, to 1.06, 0.52, 1.18 %, employing the selected conditions.