Node localization is a critical technique for acquiring location information and has emerged as a fundamental technology within wireless sensor networks (WSNs). Localization accuracy in wireless sensor networks (WSNs) can deteriorate due to uncertainties in the transmit power (TP) and path loss exponent (PLE). To address this challenge, a coarse-to-fine third-order localization method (CFTL) is proposed. First, TP uncertainty is mitigated using differential forms. The problem is then reformulated into a natural constant-based least squares estimation (NC-LSE) framework through first-order Taylor expansion and logarithmic transformations, with coarse-grained positions obtained via a linear unbiased estimation method. Second, an optimization function with PLE as the variable is constructed, and the puma optimization (PO) algorithm is employed to estimate the PLE. Third, the optimized PLE is incorporated into the differential-based generalized trust region subproblem (DGTRS) framework, and the fine-grained position is calculated using the bisection method. Additionally, the generalized inverse theorem for block matrices is applied to derive the Cram-r-Rao lower bound (CRLB) under dual-parameter uncertainty, assessing the algorithm’s effectiveness. Simulation and practical results demonstrate that the proposed method enhances localization accuracy by at least 10.96% and up to 32.18% compared to existing methods across various conditions.