Spectroscopy is regarded as a quick and nondestructive method to classify and analyze quantitatively many of elements of the soil. Visible and near-infrared re?ectance spectroscopy offers a conductive tool for investigating soil heavymetal pollution. In this work, 51 soil samples with depths of 0～10 cm were collected, which were in the Eastern Junggar coal-field mining area, Xinjiang. The soil organic matter (SOM) content, Arsenic (As) content and indoor hyperspectra were measured in the laboratory. The significant relationship between As content and hyperspectral data was conductive analysis of NPDIs, which were calculated from Vis-NIR region. For calculating the indices, on the basis of the raw spectral reflectance (R), its three mathematical transformations were calculated, i. e., the reciprocal (1/R), logarithm (lgR) and root mean square method (sqrt-R/R), respectively. The two band combination of optimized indices software V1.0 (No: 2018R11S177501, independently developed based on the JAVA) was used during the calculation of the indices. NPDIs were calculated using all possible combinations of available bands (i nm and j nm) in the full spectral region (400～2 400 nm). In the optimal spectral indices (|r|≥0.73 and p=0.001), an index of VIP≥1 was further selected as a model independent variable by the Variable importance in projection (VIP) selection method. The main goal of this work is to obtain optimized spectral index (NPDI) related to soil heavy metal As, to estimate As concentration in soil based on geographically weighted regression (GWR) model, and to investigate the plausibility of using optimized spectral index for hyperspectral detection of heavy metal Arsenic in soil of coal mining areas. To assess the performance of the soil heavy metal contents prediction models, four cross-validation metrics were used; Residual Prediction Deviation (RPD), the Coefficient of Determination (R2), the Root Mean Square Error (RMSE) and Akaike Information Criterion (ACI). The results of this study are as follows: (1) As has the largest dispersion in the study area, SOM contents in all samples are less than 2%, and the As concentration has no significant correlation with the SOM content at a significance level of 0.01 (|r|=0.113). (2) Single-bandreflectance shows low correlation with As contents, lower than 0.228. However, the highest correlation coefficient and lowest p-values (｜ｒ｜≥0.73 and p=0.001) between As and NPDIs calculated by original and transformed reflectance (R, 1/R, lgR, R) are found in theNear-infrared (NIR, 780～1 100 nm) and Shortwave-infrared (SWIR, 1 100～1 935 nm) long wavelength infrared. The original spectral region formed with long wave length near-infrared (LW-NIR) regions show highest correlation with As contents (｜ｒ｜=0.74). (3) VIP value of NPDIR(1 417/1 246), NPDI1/R(799/953, 825/947), NPDIsqrt-R(1 023/1 257, 1 008/1 249, 1 021/1 250, 1 020/1 247) and NPDIlgR(801/953, 811/953, 817/951, 825/947, 828/945) higher than 1, thus these NPDIs are chosen as independent variables. (4) From the four prediction model (GWR) performances it can be seen, the Model-a (R) showed superior performance to other three models (Model-b (1/R), Model-c (R) and Model-d (lgR)), and it has the highest validation coefficients (R2=0.831, RMSE=4.912 μg·g-1, RPD=2.321) and lowest AIC value (AIC=179.96). The hyperspectral optimized index NPDIR(1 417/1 246) may help to quickly and accurately evaluate Arsenic contents in soil, furthermore, the results provide theoretical and data support to accesse the distribution of heavy metal pollution in surface soil, promoting fast and efficient investigation of mining environment pollution and sustainable development of ecology.