We present a new lossy technique of runlength coding applied to gray-level images. In lossless runlength coding applied to gray-level images, one finds runs of gray levels that have the same value along a horizontal row and then encodes them in the form (C, L), where C is gray-level value and L is the runlength. The drawback to this approach is that it does not give much compression, as L is usually small. In our approach we relax the criterion of requiring runs of exactly the same gray level; instead we find runs of pixel gray-level values that lie within a dynamic range, the dynamic window range, and encode them in the form (C, L ). Here, C represents approximations of the pixel gray levels and L the runlength. This approach, although lossy, promises to give better compression. Additionally, if lossless Huffman coding is used in the final stage a slightly higher compression ratio is achieved. This technique was used to process a set of images and the results based on windowlength, compression ratio, root-mean-square error, and peak signal-to-noise ratio are tabulated and plotted. The compressed images and the corresponding error images are provided to illustrate the results. Additionally, performance of this algorithm was compared with the runlength coding of comparable coarsely quantized images.