High-throughput, automation-friendly and therapeutically-predictive assays are needed in early drug discovery in order to prioritise compounds and reduce the risk of new drugs causing Drug-Induced Liver Injury (DILI). We evaluated the suitability of high-throughput 3D liver spheroid models of HepG2 (C3A clone) and HepaRG cell lines to predict DILI in early drug development. Spheroids were formed in 384-well ultra-low-attachment plates and dosed via direct acoustic droplet ejection at nine half-log spaced concentrations per compound. Spheroid viability was quantified with an ATP endpoint after a 4-day incubation with 150 drugs with known DILI liability. We derived a margin of safety for each cell line defined as the ratio between the IC50 values generated for each compound to their maximum plasma concentration Cmax which resulted in optimal classification accuracy. The margin of safety can be used to estimate a maximum safe Cmax for compounds in early drug discovery for which Cmax is not yet known. Both cell lines had similar level of accuracy in predicting DILI, with HepG2 spheroids being more sensitive. HepG2 spheroids had a sensitivity of 58% and a specificity of 83%, while HepaRG spheroids had a sensitivity of 47% and specificity of 86%. Ninety-nine of the 150 compounds were used to compare the relative sensitivities of HepG2 and HepaRG spheroids. HepaRG spheroids were more sensitive to 7 compounds and HepG2 spheroids were more sensitive to 34 compounds. In conclusion, across a diverse group of drugs HepG2 spheroids were more predictive of DILI compared to HepaRG spheroids.