To explore the validity and applicability of the estimation of soil arsenic (As) content based on spectral indices, 42 soil samples were collected from a farmland in Hebei Province, China. The reflectance spectra and As content were respectively determined by using a PSR-3500 portable ground spectrometer and Inductively Coupled Plasma Atomic Emission Spectrometry. The chlorophyll index (CI), difference index (DI), sum index (SI), ratio index (RI) and simple normalized difference spectral indices (NDI and NPDI) were calculated based onlaboratory spectra, field spectra, and the direct standardization (DS) transferred field spectra. Random forest regression (RFR) models were used to estimate the soil As values using the strongly correlated spectral indices, and indices were evaluated according to the modeling accuracy. Compared with thecharacteristic absorption bands of typical soil components, the internal mechanism of spectral indices improving the inversion accuracy of soil As content was analyzed. The results show that the spectral indices method significantly enhances the correlation between spectra data and As content by combining some low-correlation band information. When compared with the full-band RFR model, the spectral indices method increased the R2p and RPD from 0. 243 and 1. 2 to 0. 730 and 2. 009, 0. 264 and 1. 213 to 0. 669 and 1. 809, 0. 334 and 1. 279 to 0. 678 and 1. 841 in the lab spectra, field spectra, and field-DS spectra respectively, and CI has the best comprehensive performance (R2p>0. 66 and RPD>1. 8). However, some of the exponential characteristic bands of the optimal spectra indices lack interpretability and cannot reveal the band combination rules for exponentially amplifying effective information and eliminating noise. The research results can provide a scientific basis for estimating heavy-metal contamination in soil using remote sensing spectroscopy based on spectral indices and even the band design of satellite payloads.