We investigate the effect of modeling tropospheric delay gradients on the station position estimates using the Japanese nationwide GPS array. The time series shows spatially coherent temporal fluctuations due to the variability of water vapor distribution. This makes it difficult to identify small crustal deformation signals in GPS time series. Precision and accuracy of station positions are known to be improved by modeling the tropospheric delay gradient, and past studies suggest that delay gradient estimates agree well with the collocated water vapor radiometer measurements. Here we pick up two intervals as long as 2 weeks in 1996 summer, when remarkable tropospheric delay gradients are expected, and investigate various influences of tropospheric delay gradients on station position estimates. First, we study spatial patterns of station position deviations caused by azimuthal asymmetry of water vapor distributions from two solutions, i.e., with and without the tropospheric delay gradient model. Second, we compare them with the estimated delay gradients. Site coordinate deviations in the solution without the delay gradient model are negatively correlated with the estimated delay gradients, but such systematic deviations disappear by introducing the delay gradient model. We found that the improvement in position accuracy is significant not only horizontally but also vertically over both of the two time intervals.