In a method for compressing an image, source data is converted into a series of blocks each containing one or more colors. A baseline target block size is determined based upon a target compression ratio for the source data. For each block in turn, a compression mode is selected from mode families, where the mode families include an n-color mode family, an interpolated mode family, a raw mode family, a BTC-VQ mode family, a lossless mode family, and a 4×4 transform mode family. In addition, for each block in turn, the block is compressed using the selected compression mode to yield a corresponding compressed block.