Haze degrades the quality of optical remote sensing data and reduces the accuracy of interpretation and classification. In this paper, we present an empirical single image-based haze removal method. Our work relies on an additive haze model, which describes at-satellite acquired radiance as the sum of globally constant path radiance, surface reflected radiance and spatially varying haze contribution. First, we search a tagged haze thickness map in band blue, green, red and near-infrared. Then a proportional strategy is adopted to infer haze thickness map of selected starting band from tagged haze thickness map and gain haze thickness map of other spectral bands from the starting band. Finally, haze is removed through subtracting haze thickness map from each band. The method is applicable for medium- and high- resolution satellite images and can handle scenes affected by different haze depths. Dehazed results are in good quality with improved visibility, enhanced image information and highly spectral consistency. The algorithm is simple enough and requires no human intervention, making it possible for use of non-experts and implementation of automatic batch processing.