| The Zariski decompostion of a pseudoeffective divisor D on a smooth surface consists in writing D as sum of a nef divisor and an effective divisor satisfying some intersection properties. Such a decomposition always exists. In higher dimension there exist several generalizations of the Zariski decomposition, one of whom is the Fujita-Zariksi decomposition. A pseudoeffective divisor admits a Fujita-Zariski decomposition if it is the sum of a nef divisor P and en effective divisor N and P is the maxiaml nef divisor such that D-P is effective. In this talk we will prove that, if a pseudoeffective divisor on a smooth projective threefold X has dimished base locus closed and of codimension 2, then its pull-back on a suitable birational model of X admits a Fujita-Zariski decomposition. |