Background and Objectives <p>Levofloxacin is an essential drug in multidrug-resistant tuberculosis (MDR-TB) treatment, but high pharmacokinetic variability necessitates therapeutic drug monitoring (TDM) to optimise exposure and improve outcomes. Traditional TDM requires invasive blood sampling, limiting feasibility. Saliva sampling, a non-invasive matrix may simplify implementation. We aimed to develop a levofloxacin population pharmacokinetic (popPK) model integrating plasma and saliva data, and to evaluate saliva-based limited sampling strategies to support model-informed TDM for levofloxacin in practice.</p> Methods <p>Adults receiving oral levofloxacin for ≥ 7 days had plasma and saliva samples collected at 0, 2, and 5 h post-dose. A plasma-and-saliva popPK model was developed using NONMEM, including covariate evaluation. Predictive performance of saliva-based limited sampling strategies for estimating plasma AUC₂₄ was assessed using Bayesian estimation and Monte Carlo simulations.</p> Results <p>A total of 57 patients with 342 paired plasma-saliva samples were evaluated. One-compartment model incorporating saliva bio-compartment with first-order absorption (with a lag time) and elimination best described the data. No significant covariates were identified. Simulations showed that the three-point saliva strategy (0, 2, 5 h) predicted plasma AUC₂₄ within clinically acceptable limit (&lt; 15% prediction error). A single 2-h sample yielded comparable accuracy. Two- and three-sample saliva strategies up to 16 h post-dose also showed clinically acceptable accuracy.</p> Conclusions <p>The developed popPK model enables reliable estimation of levofloxacin exposure from limited saliva samples. A single 2-h post-dose saliva concentration may support model-informed levofloxacin TDM in routine MDR-TB care. This approach may be beneficial in settings where plasma-based TDM is logistically or ethically challenging.</p>