We have studied three methods of spatially interpolating GPS-estimated wet delay time series. Five months of data from three sites in the Swedish continuously operating GPS network were used for interpolation to a fourth site. Different geometries were used, resulting in interpolation distances of 40--120 km. To assess the quality of the interpolations we compared the obtained time series with estimated wet delay from a GPS station and with radiosonde data. A method based on the theory of atmospheric turbulence showed the lowest errors when compared to both the GPS estimates and the radiosonde data. The rms errors using this method are typically 1 cm. ¿ 1998 American Geophysical Union