An adaptive coding scheme for digital communication over time-varying channels is presented. The scheme is based on a finite-state Markov channel model. Emphasis is on the adaptation of the error protection to the actual channel state. The throughput gains that are achieved by the adaptive scheme relative to the conventional nonadaptive coding methods are demonstrated by several examples. Of special interest is the use of punctured convolutional codes with maximum-likelihood Viterbi algorithm to enable adaptive encoding and decoding without modifying the basic structure of the encoder and the decoder.< >