As an octane enhancer in commercial gasoline, ethanol is widely used by its oxygenated and renewable natures. In this study, five gasoline-ethanol blended fuels are studied with anti-knock index of 87-92: (1) Haltermann CARB LEV III E10 gasoline, (2) FACE C gasoline with 7% ethanol (by vol), (3) FACE C with 14% ethanol (by vol), (4) FACE J with 23% ethanol (by vol), (5) FACE J with 36% ethanol (by vol). First, surrogate models are formulated for the five fuels. By conducting a Jacobian-aided directed relation graph with error propagation (DRGEP) approach, a much improved performance is shown over the original counterpart for the multiple heavy fuel surrogate mechanism reduction. Based on the analysis, skeletal mechanisms generated after a hump region with an acceptable error in graph-based approaches is shown to deteriorate the further reduction potential. Starting from a detailed 1134-species mechanism, a 130-species skeletal mechanism and an 80-species reduced mechanism are finally developed with high fidelity under gasoline compression ignition (GCI) engine conditions, combining the DRGEP, target search algorithm (TSA) and quasi-steady state assumption (QSSA) approaches. Additionally, the low temperature heat releases in homogeneous charge compression ignition (HCCI) engine are validated under wide ranges of intake temperatures and pressures.