Terahertz spectroscopic measurements under ordinary atmospheric conditions may suffer interferences from water vapor absorption in the ambient air. A manifold of narrow absorption lines appears in the terahertz spectrum at particular frequencies corresponding to the pure rotational transitions of water molecules. For real-world data, such effect results in unwanted spectral artifacts in the deconvolved spectrum of the examined sample and thus complicates its frequency-dependent characterization. In this paper we use a signal postprocessing algorithm consisting of line shape fitting and spectral subtraction procedures to eliminate the water lines. Restoration of terahertz signals from simulated data and low-humidity measurements is first demonstrated to validate the algorithm. Furthermore, to overcome the difficulty of eliminating strong lines which lead to possible excessive absorption under high-humidity environment, we propose to modify the objective function in spectral subtraction by smoothing the residual spectrum to get acceptable performance.