Monitoring soil salinity is essential for agricultural development and ecological systems in the coastal saline area. The global navigation satellite system (GNSS) interferometry reflectometry (GNSS-IR) can provide new opportunities to retrieve long-term soil salinity at the point-scale theoretically, because the utilized L-band of GNSS-IR is sensitive to soil dielectric permittivity, while the soil salinity can affect its imaginary part. However, the method of soil salinity retrieval has not been researched currently. This study, taking the GNSS-IR data from the station located in the coastal saline area, gives the first evaluation of using this data source for soil salinity retrievals. First, the three interferogram metrics (i.e., phase, amplitude, and reflector height) and their corresponding statistics are extended to include the contributions from the environmental conditions. Then, the model for semi-empirically retrieving soil salinity from these parameters is constructed through the gradient boosting regression tree (GBRT) machine learning (ML) technique. The results show that the phase and height and their corresponding statistics have a relatively strong relationship with the soil salinity as independent variables. Meanwhile, the soil salinity retrieved from the global positioning system and BeiDou navigation satellite system (BDS) agree and correlate well with the in-situ measurements derived from the 5TE sensor ( R varies from 0.670 to 0.808, RMSE varies from 1.350 to 1.895 mS/cm, and MAE ranges from 1.049 to 1.749 mS/cm). The work shows the capability of GNSS-IR in retrieving soil salinity and considerably increases the service available from the geodetic-grade GNSS receivers.