MRI- and MRS-based proton density fat fraction (PDFF) techniques require accurate modeling of the multi-peak spectrum of triglycerides (TG) in order to achieve accurate hepatic fat quantification. However, variations in TG spectrum may lead to quantification variability. We performed a secondary analysis of adults with biopsy-confirmed nonalcoholic steatohepatitis undergoing confounder-corrected chemical-shift-encoded 3T MRI and MRS, and calculated variant PDFF values using a range of biologically plausible spectral models. Within the range of fat fractions seen in the liver, PDFF estimation using MRI and MRS was robust to variability in the TG spectrum. Greater bias was seen when the baseline fat fraction was higher, but remained low.