Quantification of in vivo proton magnetic resonance spectra (1H-MRS) still commonly involves evaluation of exclusively the real part of acquired spectral signals, but ignoring the information contained in the imaginary component may limit precise identification of individual metabolite contributions. Here, we assess quantification precision relative to SNR and noise correlation for both real and complex linear combination model fits of simulated 1H-MRS spectra reflecting brain metabolite concentrations and T2. Extending our results to inclusion of measured in vivo baselines, we demonstrate consistent improvements in metabolite quantification precision and/or accuracy by complex relative to real fits.