The effect of troposphere on the signals emitted from global navigation satellite system (GNSS) satellites, appears as an extra delay in the measurement of the signal traveling from the satellite to receiver. This delay depends on the temperature, pressure, humidity as well as the transmitter and receiver antennas location. In GNSS positioning, tropospheric delay effects on accuracy of different components of obtained coordinates. In RTK networks the amount of this parameter is determined by solving double difference observation equations between reference stations and then is interpolated for rover receiver. Tropospheric delay consists of a wet part and a dry part. The dry part that forms about 90 percent of total delay, is related to station height. So in the cases that the height of rover station is significantly different from the average height of reference stations, reduction in accuracy of interpolation is expectable. To investigating this issue, in this article we compared interpolation accuracy of double difference tropospheric delay in two networks with different structure. In both of networks, we have a central receiver that is surrounded with four other receivers. We considered the central receiver as rover station and the others as reference stations. The main difference between these networks is about stations height. In the first network that is named Sima, the difference between the height of rover station and average height of reference stations is 122 meters. The amount of this parameter is 1095 meters for the second network that is named Ebry. To comparing the accuracy of tropospheric delay interpolation in these networks, we determined zenith tropospheric delays (ZTD) for all stations by processing GNSS observations using CSRS-PPP (Canadian Spatial Reference System – Precise Point Positioning) online service. Then we selected the nearest reference station to rover as master reference station. In the following we identified the satellites that were visible in 100 epochs for all stations. Between these satellites, one of them with the most elevation angle was selected as reference satellite. ZTD’s were converted to slant tropospheric delay in satellite-receiver direction using global mapping function. Then double differenced tropospheric delays between the reference satellite and the others and between the master reference station and other reference stations, were determined. Finally this parameter was computed for the position of rover station using interpolation with a two parameter linear equation. After computing RMSE (Root Mean Square Error) of interpolated values, we found that the accuracy of interpolation decreased significantly in the second network. Therefore we can conclude that the difference between the height of rover station and the height of reference stations, has a direct effect on accuracy of tropospheric delay interpolation in RTK networks. So in the following of the article, we introduced a new method to eliminate height variations effects on interpolation accuracy of tropospheric delay. After using this method RMSE of interpolation decreases from 32 mm to 9 mm in the first network and in the second network decreases from 228 mm to 14 mm. in other words we have 69.2 and 93.7 percent of accuracy improvement in these networks. Due to these results, we expect a positive effect on positioning accuracy by applying this method in RTK networks.