We have used a ground-based microwave radiometer, known as a water vapor radiometer, to investigate the local spatial and temporal variation of the wet propagation delay for a site on the west coast of Sweden. The data were obtained from a wide range of azimuths and from elevation angles greater than 23.6¿ (air mass 2.5). Visual inspection of the data suggested a simple ''cosine azimuth'' variation, implying that a first-order gradient model was required. This model was adequate for short time spans up to approximately 15 min, but significant temporal variations in the gradient suggested to us that we include gradient rate terms. The resulting six-parameter model has proven adequate (rms delay residual ~1 mm) for up to 30 min of data. Assuming a simple exponential profile for the wet refractivity gradient, the estimated gradient parameters imply average surface wet-refractivity horizontal gradients of order of 0.1--1 N km-1. These gradients are larger, by 1--2 orders of magnitude, than gradients determined by others by averaging over long (~100-km) distances. This result implies that for applications that are sensitive to local gradients, such as wet propagation-delay models for radio-interferometric geodetic studies, the use of meteorological data from widely spread stations may be inadequate. The gradient model presented here is inadequate for times longer than about 30 min, even if no gradients are present, because of the complicated stochastic like temporal behavior of the wet atmosphere. When gradients are present, they can change magnitude by ~50% over 10--15 min. Nevertheless, our ability to fit the radiometer data implies that on timescales ≤30 min and for elevation angles≥23.6¿, the local structure of the wet atmosphere can be described with a simple model. (The model is not limited to this range of elevation angles in principle.) The estimated gradient and gradient rate vectors have preferred directions, which indicates a prevailing structure in the three-dimensional temperature and humidity fields, possibly related to systematic behavior in large-scale weather systems and/or the local air-land-sea interaction at this site. ¿ American Geophysical Union 1993