Vancomycin remains a key treatment for infections caused by β-lactam-resistant gram-positive cocci. While there is unanimity that only drug not bound to plasma proteins is pharmacologically active, a wide range of values and interdependencies for the unbound fraction (fu) of vancomycin have been reported in the past. In the present study, we evaluated 706 plasma samples from 228 adult in-patients who were sent for therapeutic drug monitoring. Total and free concentrations of vancomycin were analyzed by a validated method using ultrafiltration and HPLC-UV. Covariate effects on fu were assessed by a linear mixed-effects model. The mean unbound fraction was 72.2 ± 5.5% (coefficient of variation 7.7%, range 53-93%), the intra-individual and inter-individual variability were low (median coefficient of variation 5.7% and 6.4%, respectively). The unbound fraction was independent of total vancomycin, plasma albumin or total protein concentrations, or other biochemical or demographic variables, and did not differ between patients treated inside or outside of intensive care (P=0.465). Linear mixed-effects modeling confirmed low overall variability (coefficient of variation 7.0%), decomposed into 2.2% inter-individual, 3.8% inter-occasion, and 5.5% residual variability. Method comparison showed an excellent agreement between high-performance liquid chromatography with ultraviolet detection (HPLC-UV) and the immunoassay used for routine drug monitoring (bias +0.2%). Free concentrations of vancomycin can be reliably predicted from total concentrations. An unbound fraction of approximately 70% provides a robust and clinically useful estimate. Remaining variability appears to be primarily methodological, and not of clinical relevance.