An improved decision feedback equalizer for use with digital communications. The equalizer includes an error detector, a process controller, a parameter selector and a data buffer for temporarily storing a digital data signal received from a communication channel. The error detector determines whether the equalizer is accurately tracking changes in the communication channel's characteristics or is lost. When the error detector determines that the equalizer is lost, the process controller responsively generates control signals for initiating an optimal retraining/recovery method for the prevailing conditions. In some retraining/recovery methods, data is temporarily stored in the buffer. The stored data is later retrieved and processed once the equalizer is retrained. Retraining is performed using a retraining signal received via the communication channel or, if available, a portion of the data signal which is suitable for retraining, thereby permitting more rapid resumption of data reception. In addition, the parameter selector, in response to a control signal from the process controller and a signal which is indicative of the rate of change of the communication channel's characteristics, dynamically adjusts the convergence rate of the equalizer to provide to an optimum value for performing retraining or tracking changes in the channel's characteristics.