Remote sensing images (RSIs) taken in hazy conditions, such as haze, fog, thin could, snow, silt, dust, offgas, etc., suffer from sever color and contrast degradations. Dehazing algorithm is therefore highly demanded to restore hazed RSIs from their degradations. In the literatures, most dehazing algorithms were originally designed for natural images dehazing (NID). For our analysis, the physical model of NID is different from that of RSI dehazing (RSID), which was not clearly addressed yet. In this paper, a new concept of “virtual depth” concerning physical model of RSI is firstly raised. Virtual depth is different from real depth of nature image in that the former gives the distance of an object departing from the foreground, while the later measures the coverings of the earth’s surface, such as snow, dust, cloud and haze/fog. These coverings act as the hazes in a natural image, providing the hint of foreground and background. Secondly, an Iterative Dehazing for Remote Sensing image (IDeRS) is proposed, in which dehazing operator is implemented iteratively to remove haze progressively until arriving at a satisfied result. In IDeRS, we also raise a fusion model for combining patch-wise and pixel-wise dehazing operators to overcome both halos and over-saturation caused by them respectively. Extensive experimental results tested on publicly available databases demonstrate that the proposed IDeRS outperforms most state-of-the-arts in RSID.