To test the hypothesis that alternative splicing could be an adaptive mechanism for populations subject to multi-generational estrogenic exposures, we compared estrogen receptor alpha (ERα) splicing variants in two populations of killifish (Fundulus heteroclitus): one resident in an estrogenic polluted environment (New Bedford Harbor, NBH, MA, USA) and one from a relatively uncontaminated reference site (Scorton Creek, SC, MA, USA). In total we identified nineteen ERα variants, each with deletions of one or more coding exons. Four of the variants with potential functional relevance were analyzed by qPCR to test for population differences in expression by tissue type, site, sex, seasonal reproductive status and estrogen treatment. Significantly, a 5′-truncated short form variant (ERαS) was highly expressed in liver and ovary, and was associated with seasonal reproductive activity in SC but not NBH fish. Both ERαS and the full-length long variant (ERαL) were estrogen-inducible (ERαS > ERαL) but the induction response was lower in NBH than in SC fish. In contrast, NBH killifish were hyper-responsive to estrogen as measured by expression of two other estrogen responsive genes: vitellogenin (Vtg) and aromatase B (AroB). Most strikingly, two ERα deletion variants (Δ6 and Δ6–8), lacking ligand binding and activation function domains, were identified in a subset of NBH fish, where they were associated with reduced responsiveness to estrogen treatment. Together, these results support the hypothesis that alternative splicing of the esr1 gene of killifish could be an autoregulatory mechanism by which estrogen modulates the differential expression of ERα, and suggests a novel and adaptive mechanistic response to xenoestrogenic exposure.