Method for predicting properties of a cellulose-fibre-based product, a suspension of cellulose fibre being analysed during manufacture of the product by means of spectroscopic measurements (13, 15) in a selected spectrum in the wavelength range 200-25000 nm. According to the invention, each sample quantity is diluted (3, 4) prior to said analysis, following which one partial flow (6) is dewatered and dried (8) and used for said spectroscopic measurements (13, 15), while a second partial flow (5) of the diluted sample quantity is used for analysis (7, 14) of physical fibre data, by means of image analysis, each sample quantity analysed generating at least 100 data points from the spectroscopic measurement, and at least 50 data points with regard to said physical fibre data, in the form of one or more physical fibre property distributions, which data points are combined in multivariate data processing (16), for said prediction.