A general method is proposed to choose, among several benchmarks, suitable assimilation database parameters to be used in adjustments.All these benchmarks are predetermined keeping in mind specific goals of the data assimilation.The main principle of the novel methodol. is that of excluding groups of strongly cross-correlated computed parameters by estimating their posterior values of χ2.The assimilation tries minimizing the mean χ2 for the entirety of the considered benchmarks, including those target parameters which are not part of the selected assimilation database.The methodol. currently applied to the JEFF-3.3 library, is demonstrated by adjusting data for 235U, 238U, 239Pu and 240Pu in conjunction with a class of integral parameters (effective multiplication factors) of fast criticalities including the Godiva and Jezebel spheres, for which the effect of varying resonance parameters in the calculations may be largely neglected.The computed mean χ2 of 36 values in total, among which seven refer to the assimilation database, is halved, and the variance of the individual χ2-values decreases by a factor of 6 with respect to the prior values.More explicitly, the posterior values of resp. 1.10 and 3.77 for χ2 and the variance of the individual χ2-values are much smaller than the corresponding prior values of 2.32 and 22.38, which is significant.As regards the adjustment, for 235U, it is found that μ and the inelastic scattering cross-section resp. decrease by maximum 3% and 2.6%; ν is uniformly decreased by maximum 0.053% and the fission spectrum is somewhat harder.For 238U, μ increases by maximum 5.4% and the inelastic scattering cross-section is lowered by 1%.For 239Pu, ν is uniformly increased by maximum 0.027% and the fission spectrum is slightly softer.For 240Pu, the fission cross-section is decreased by maximum 3.9%.The proposed adjustments are well within the uncertainty limit of one standard deviation for the nuclear data.