Fast and reliable methods for in situ monitoring of soil NO3–N concentration could help guide efforts to reduce NO3–N losses to ground and surface waters from agricultural systems. While several studies have been done to indirectly estimate NO3–N concentrations from time domain spectra, no research has been conducted using a frequency domain technique. Hence, the goal of this laboratory study was to estimate NO3–N concentrations from frequency-response data obtained in a frequency range of 5 Hz to 13 MHz. Dielectric spectra of soil samples wetted to five different volumetric water contents (VWC) with 24 solutions containing different concentrations of KNO3 and KCl were analyzed using a partial least squares (PLS) regression method. Global models could not estimate NO3–N concentrations with sufficient accuracy. Models based on the imaginary part of the permittivity were better than those based on the real part. The PLS model estimates were improved when low VWC data and high Cl− concentration were eliminated, reducing the RMSE for NO3–N from 57 to 28 mg L−1. The best results were obtained when the PLS models were constructed at fixed VWC levels using the data without high Cl− concentration. The performance of these models improved with increasing VWC level, reaching the lowest RMSE of 18 mg L−1 at VWC of 0.30 m3 m−3.