The ENDF/B-IV prescription fails to represent correctly the 238U total (and scattering) cross section between the levels of the resolved range. We show how this representation can be improved by properly accounting for the contribution of levels outside the resolved region to the cross section at energies inside the resolved region, and by substituting the more precise multi-level Breit-Wigner formula for the presently used single-level formula. We illustrate the importance of computing accurately the minima in the total cross section by comparing values of the self-shielded capture resonance integral computed with ENDF/B-IV and with a more accurate cross-section model.