Abstract: High-precision gold resource exploration provides critical data support for mine design and production planning, while the analysis of ore containing coarse-grained gold, low-grade and polymetallic associations remains a current industry challenge. Natural gold in such ores is highly malleable and unevenly distributed, leading to severe sampling representativeness issues with conventional sample preparation. Moreover, common covering agents used in existing fire-assay methods (e.g., sodium carbonate-borax, sodium chloride) suffer from weak protection, high-temperature loss of gold due to volatilization, and release of toxic gases, making it difficult to balance analytical accuracy with environmental friendliness. To address the three core problems of insufficient representativeness of coarse-grained gold, gold loss during fusion, and environmental pollution, this study developed an integrated analytical system of “gravity preconcentration–graded detection–composite-cover fire assay” using typical coarse-grained gold ore from Muli, Sichuan, and multi-grade certified reference materials. Gravity preconcentration was achieved by sieving at 0.074 mm (200 mesh), with the oversize heavy minerals subjected to total analysis and the undersize light minerals analyzed in parallel; gold content was calculated by mass-weighted averaging, ensuring sampling representability from the source. Through systematic screening and optimization, an environmentally friendly composite covering agent of anhydrous sodium sulfate-sodium chloride (4:1) was developed. This covering agent forms a stable, dense molten layer at high temperature, and its excellent film-forming ability and chemical inertness effectively isolate the melt from air, significantly suppressing the loss of ultrafine gold particles (＜1 μm) by gas-flow entrainment and material splashing, while being cost-effective and environmentally benign. The relative error of this method for gold certified reference materials ranged from −6.25% to 1.05%, and the spike recovery of gold was 91.2%–104.3%; for actual samples, the spike recovery of gold was 93.5%–102.5%. F-test and t-test (α = 0.05) confirmed that the precision and trueness of the proposed method are superior to those of the conventional fire-assay method (GB/T 20899.1-2019). The method detection limit is 0.03 μg/g, the quantification limit is 0.09 μg/g, and it is applicable to a wide gold grade range of 0.1–130 μg/g. Through whole-process optimization and coupling of pretreatment and detection steps, this study provides a reliable technical solution for green, high-precision analysis of complex coarse-grained gold-bearing ores.