While sulfur isotopes have proven powerful tracers of sulfur (re)cycling within subduction zones, the origin of the 34S-enrichment seen in arc magmas remains a subject of debate, with competing hypotheses implicating both mantle and crustal processes. Herein, we investigate these competing models through the study of the Early–Middle Jurassic Talkeetna Arc section exposed in the Chugach Mountains, south-central Alaska, reporting the sulfur isotope systematics of rocks spanning a depth transect that extends from the upper mantle (∼0.9–1.2 GPa, ∼30–35 km) up into the mid-crust (∼0.2–0.5 GPa, ∼5–15 km). Marked by an increase in δ34S values from −1.28 to +5.61‰, these data reveal a significant isotope effect associated with melt differentiation, with 32S being preferentially sequestered from primitive basaltic melts into ultramafic and mafic cumulates, yielding progressively more evolved melts enriched in 34S. Leveraging a quantitative petrological model that constrains the liquid line of descent and sulfide saturation as a function of sulfur content, oxygen fugacity, pressure and temperature, we then demonstrate that, in the presence of dissolved oxidized sulfur species, the saturation of immiscible magmatic sulfides is capable of generating the observed sulfur isotope fractionation. While fluid saturation is unlikely to occur at lower crustal depths, ascending melts will almost certainly experience fluid saturation and degassing. As a result, sulfide immiscibility and magma degassing are not mutually exclusive and may, instead, represent complementary processes that combine to explain the observed range of positive δ34S compositions. In Talkeetna, crustal assimilation is unlikely to have played a role to increase δ34S values as there was little to no assimilation of pre-existing oceanic crust containing seawater sulfate during the formation of the Talkeetna Arc section. Furthermore, our model suggests that primitive melt in isotopic equilibrium with the most primitive ultramafic cumulates have mantle-like δ34S values between ∼0 and +1.1‰, which suggest limited input of 34S-enriched slab-derived sulfur into the sub-arc mantle. Collectively, these findings emphasize that mantle-like δ34S values for primitive arc melt may be common, while revealing the underappreciated importance of deep crustal crystallization of immiscible magmatic sulfides to generate positive δ34S compositions in erupted arc magmas.