A novel method of developing test patterns for digital video coding systems is presented. This method is illustrated for an MPEG-2 encoder where, subject to fixed encoding parameters, the input frames are identically the expected output frames of reconstructed video. The method is based upon a method of developing test bitstreams for MPEG decoders that is also described. Both methods rely upon bit-accurate modeling of the system under test by a software codec, usually a C program. Natural video may be used to test digital video systems; however, encoded natural images seldom provide adequate coverage of either fixed-length or variable- length binary codeword spaces. Test bitstreams may be readily constructed for decoders to cover specific codeword subspaces, but they decode as noticeably artificial video frames. 'Artificial' test patterns are obtained for an encoder by first decoding a specially constructed test bitstream, then holding its higher-level encoding parameters fixed and iteratively encoding the 'artificial' video frames until convergence is reached, i.e., input frames equal output frames. Convergence occurs after a few or dozens of iterations, and the resulting 'artificial' test patterns retain substantial coverage.