We study deep-inelastic scattering factorization on a nucleon in the end-point regime x B ∼ 1 − O(Λ QCD /Q) where the traditional operator product expansion is supposed to fail. We argue, nevertheless, that the standard result holds to leading order in 1−x B due to the absence of the scale dependence on (1 − x B )Q. Refactorization of the scale (1 − x B )Q 2 in the coefficient function can be made in the soft-collinear effective theory and remains valid in the end-point regime. On the other hand, the traditional refactorization approach introduces the spurious scale (1 − x B )Q in various factors, which drives them nonperturbative in the region of our interest. We show how to improve the situation by introducing a rapidity cut-off scheme, and how to recover the effective theory refactorization by choosing appropriately the cut-off parameter. Through a one-loop calculation, we demonstrate explicitly that the proper soft subtractions must be made in the collinear matrix elements to avoid double counting.