The soil matrix can sequester arsenic (As) and reduces its exposure by soil ingestion. In vivo dosing studies and in vitro gastrointestinal (IVG) methods have been used to predict relative bioavailable (RBA) As. Originally, the Ohio State University (OSU-IVG) method predicted RBA As for soils exclusively from mining and smelting sites with a median of 5,636 mg As kg. The objectives of the current study were to (i) evaluate the ability of the OSU-IVG method to predict RBA As for As contaminated soils with a wider range of As content and As contaminant sources, and (ii) evaluate a modified extraction procedure's ability to improve prediction of RBA As. In vitro bioaccessible (IVBA) by OSU-IVG and California Bioaccessibility Method (CAB) methods, RBA As, speciation, and properties of 33 As contaminated soils were determined. Total As ranged from 162 to 12,483 mg kg with a median of 73 mg kg. RBA As ranged from 1.30 to 60.0% and OSU-IVG IVBA As ranged from 0.80 to 52.3%. Arsenic speciation was predominantly As(V) adsorbed to hydrous ferric oxide (HFO) or iron (Fe), manganese (Mn), and aluminum (Al) oxides. The OSU-IVG often extracted significantly less As in vitro than in vivo RBA As, in particularly for soils from historical gold mining. The CAB method, which is a modified OSU-IVG method extracted more As than OSU-IVG for most soils, resulting in a more accurate predictor than OSU-IVG, especially for low to moderately contaminated soils (<1,500 mg As kg) with RBA As = 0.81 IVBA As + 3.2, r = 0.91.