In [LP] the authors defined symplectic "Local Gromov-Witten invariants" associated to spin curves and showed that the GW invariants of a Kähler surface X with p g > 0 are a sum of such local GW invariants. This paper describes how the local GW invariants arise from an obstruction bundle (in the sense of Taubes [T]) over the space of stable maps into curves. Together with the results of [LP], this reduces the calculation of the GW invariants of complex surfaces to computations in the GW theory of curves. * partially supported by the N.S.F.where (i k ) * is the induced map from the inclusion D k ⊂ X. Thus the computation of the GW invariants of Kähler surfaces with p g > 0 is reduced to the problem of calculating the local invariants of spin curves.Recently, Kiem and Li [KL] defined the local invariants by algebraic geometry methods and proved the formulas for degree 1 and 2 local invariants conjectured by Maulik and Pandharipande [MP]. The first author [Lee2] also reproved those formulas by adapting the symplectic sum formula of [IP2] to local GW invariants.Because (0.1) applies to all GW invariants, not just those of the "embedded genus", one cannot apply Seiberg-Witten theory. Nor can they be computed by the usual methods of algebraic geometry, such as localization and Grothendieck-Riemann-Roch, because the linearized J α -holomorphic map equation is not complex linear. In particular, when genus(D k ) > 0 the local invariants in (0.1) are not the same as the "local GW invariants" used to study Calabi-Yau 3-folds [BP] or the "twisted GW invariants" defined by Givental.While the local invariants are defined in terms of the GW theory of the (complex) surface N D one would like, as a step toward computation, to recast them in terms of the much betterunderstood GW theory of curves (cf. [OP]). This paper uses geometric analysis arguments to prove that the local GW invariants of a spin curve (D, N ) arise from a cycle in the space of stable maps into the curve D. The cycle is defined by constructing an "obstruction bundle". While the basic idea is clear and intuitive, the construction is difficult because of technical issues involving the construction of a complete space of maps.The intuition goes like this. The tautological 2-form α on N D determines an almost complex structure J a . By the Image Localization Property, the space of stable J α -holomorphic maps into N D representing d[D] is the same as the space of degree d stable maps into D: M Jα g,n (N D , d[D]) = M g,n (D, d).(0.2)Counting dimensions, one sees that the formal dimension of M g,n (D, d) is exactly twice the dimension of the virtual fundamental class that defines the local GW invariants of N D (when n = 0). The dimensions do not match because J α is not generic. Perturbing J α to a generic J effectively reduces the space of maps to a half-dimensional cycle in M g,n (D, d) that defines the local GW invariants of the spin curve (D, N ). To understand this reduction, we use another remarkable property of the J α -holomorphic maps:Injectivity Property: Th...