An improved low bit-rate interframe video encoder is disclosed of the type known as a hybrid coder. A hybrid coder achieves image compression by using a two-dimensional signal transformation on blocks of differential pel data in the forward loop of a DPCM coder. The transform coefficients of each block are then quantized and entropy coded for transmission. Coding efficiency is in part determined by the size of the transform block. Larger blocks are more bit efficient because of the lower quantity of overhead data required, but require a complex transformer hardware implementation. In addition, larger blocks produce annoying block distortion. The disclosed coder has the advantages of both large and small block size division of the video image. In the disclosed coder, after differential combination (307) with a corresponding block from the previous frame, each m.times.m block of pel data is sub-divided (309) into smaller n.times.n (n<m) blocks of data which are individually transformed by a two-dimensional discrete cosine transformer (311). After the coefficients of each sub-block are quantized (312), the main block is reconstructed (314). An entropy encoder (315) scans the sub-blocks and codes the resultant string of scanned coefficients in such a way that the inter-sub-block correlation is efficiently exploited.