A series of 12 correction algorithms for quantitative XRF analysis have been compared. The algorithms are applied to intensities, calculated by the NRLXRF program for hypothetical binary, tertiary and one more complex alloy. The influence coefficients are determined by a least‐squares procedure. It is shown that, when both enhancement and absorption occur, at least a two ‐coefficient model is necessary and for more complex samples cross‐effect terms reduce the maximum errors substantially.