Background and purpose Blood-based biomarkers are a non-invasive solution to predict the risk of conversion of mild cognitive impairment (MCI) to dementia. The utility of free plasma amyloid peptides (not bound to plasma proteins and/or cells) as an early indicator of conversion to dementia is still debated, as the results of studies have been contradictory. In this context, we investigated whether plasma levels of the free amyloid peptides Aβ1–42 and Aβ1–40 and the free plasma Aβ1–42/Aβ1–40 ratio are associated with the conversion of MCI to dementia, in particular AD, over three years of follow-up in a subgroup of the BALTAZAR cohort. We also compared their predictive value to that of total plasma Aβ1–42 and Aβ1–40 levels and the total plasma Aβ1–42/Aβ1–40 ratio. Methods The plasma Aβ1–42 and Aβ1–40 peptide assay was performed using the INNO-BIA kit (Fujirebio Europe). Free amyloid levels (defined by the amyloid fraction directly accessible to antibodies of the assay) were obtained with the undiluted plasma, whereas total amyloid levels were obtained after the dilution of plasma (1/3) with a denaturing buffer. Free and total Aβ1–42 and Aβ1–40 levels were measured at inclusion for a subgroup of participants (N = 106) with mild cognitive impairment (MCI) from the BALTAZAR study (a large-scale longitudinal multicenter cohort with a three-year follow-up). Associations between conversion and the free/total plasma Aβ1–42 and Aβ1–40 levels and Aβ1–42/Aβ1–40 ratio were analyzed using logistic and Cox Proportional Hazards models. Demographic, clinical, cognitive (MMSE, ADL and IADL), APOE, and MRI characteristics (relative hippocampal volume) were compared using non-parametric (Mann-Whitney) or parametric (Student) tests for quantitative variables and Chi-square or Fisher exact tests for qualitative variables. Results The risk of conversion to dementia was lower for patients in the highest quartile of free plasma Aβ1–42/Aβ1–40 (≥ 25.8%) than those in the three lower quartiles: hazard ratio = 0.36 (95% confidence interval [0.15–0.87]), after adjustment for age, sex, education, and APOE ε4 (p-value = 0.022). This was comparable to the risk of conversion in the highest quartile of total plasma Aβ1–42/Aβ1–40: hazard ratio = 0.37 (95% confidence interval [0.16–0.89], p-value = 0.027). However, while patients in the highest quartile of total plasma Aβ1–42/Aβ1–40 showed higher MMSE scores and a higher hippocampal volume than patients in the three lowest quartiles of total plasma Aβ1–42/Aβ1–40, as well as normal CSF biomarker levels, the patients in the highest quartile of free plasma Aβ1–42/Aβ1–40 did not show any significant differences in MMSE scores, hippocampal volume, or CSF biomarker levels relative to the three lowest quartiles of free plasma Aβ1–42/Aβ1–40. Conclusion The free plasma Aβ1–42/Aβ1–40 ratio is associated with a risk of conversion from MCI to dementia within three years, with performance comparable to that of the total plasma Aβ1–42/Aβ1–40 ratio. Threshold levels of the free and total plasma Aβ1–42/Aβ1–40 ratio could be determined, with a 60% lower risk of conversion for patients above the threshold than those below.