This　paper　revisits　the　fuzzy　logarithmic　least　squares　method　(LLSM)　in　the　analytic　hierarchy　process　and　points　out　its　incorrectness　in　the　normalization　of　local　fuzzy　weights,　infeasibility　in　deriving　the　local　fuzzy　weights　of　a　fuzzy　comparison　matrix　when　the　lower　bound　value　of　a　non-normalized　fuzzy　weight　turns　out　to　be　greater　than　its　upper　bound　value,　uncertainty　of　local　fuzzy　weights　for　incomplete　fuzzy　comparison　matrices,　and　unreality　of　global　fuzzy　weights.　A　modified　fuzzy　LLSM,　which　is　formulated　as　a　constrained　nonlinear　optimization　model,　is　therefore　suggested　to　tackle　all　these　problems.　A　numerical　example　is　examined　to　show　the　applicability　of　the　modified　fuzzy　LLSM　and　its　advantages.　©　2006　Elsevier　B.V.　All　rights　reserved.