Disease-warning systems are an essential component of integrated pest management (IPM). Disease-warning systems assess the risk of occurrence or buildup of a crop disease based on information about the weather, disease-causing organism, and/or the crop. The most common inputs to disease-warning systems are weather data. Use of commercially available, site-specific weather estimates in disease-warning systems can reduce the cost, inconvenience, and unreliability of on-site weather monitoring, resulting in wider implementation of IPM. We validated site-specific weather estimates, provided by SkyBit, Inc. (Bellefonte, PA), for 15 sites in Iowa, Nebraska, and Illinois during April-September 1997-1999 against on-site measurements, and compared their performance in four simulated disease-warning systems. Compared to on-site measurements, SkyBit overestimated daily mean temperature by 0.2 deg C; rainfall duration by 2.6 h/day; rain occurrence (in days) by 12%; and rain amount by 2.5 mm/day. SkyBit underestimated duration of periods with relative humidity (RH)>90% by 4.2 h/day and leaf wetness duration (LWD) by 1.3 h/day. On an hourly basis, SkyBit overestimation of temperature was greater during the day than at night, and was greatest during hours with recorded wetness and RH>90%. SkyBit overestimated LWD during the day and underestimated LWD during the night. SkyBit underestimation of periods of RH>90% was greater at night than during the day. On a spatial scale, SkyBit estimated mean temperature and rain occurrence for a given site with greater accuracy than ground-station measurements made at distances of >50 km away. Accuracy of ground-station measurements of duration of RH>90% made <600 km away exceeded accuracy of mean SkyBit estimates. Accuracy of ground-station measurements of LWD made £1,300 km from a given site exceeded mean SkyBit accuracy. When a CART/SLD/Wind (Classification and Regression Tree/Stepwise Linear Discriminant/Wind) correction model was applied, accuracy of SkyBit estimation of LWD exceeded that of ground-station measurements made at >50 km from any site. In BLITECAST (potato late blight) disease-warning system simulations, SkyBit data (hours RH>90%) severely underestimated (P < 0.0001) severity values, which resulted in fungicide sprays more than two months late compared to on-site measurements of RH>90%. In potato early blight simulations, SkyBit data (mean daily temperature) estimated physiological days as accurately (P=0.13) as on-site data. Application of the CART/SLD/Wind correction model greatly improved performance of SkyBit data (LWD) in TOM-CAST (tomato early blight) warning system simulations. In simulation of a rain-dependent model predicting post-bloom fruit drop of citrus, SkyBit data (total rain during the past 5 days) resulted in higher (P < 0.0001) disease incidence, an earlier (P=0.0007) threshold for the first fungicide spray, and recommended more (P=0.008) fungicide sprays than on-site measurements. Deviations of SkyBit estimates of duration of RH > 90%, LWD, rainfall, and hourly temperature suggest a need to identify sources of error and to refine algorithms for estimation of these weather variables. The CART/SLD/Wind correction model developed at Iowa State University greatly enhances accuracy of SkyBit estimation of LWD.