In this paper we derive a computationally efficient algorithm for accurately approximatinga long FIR filter by a reduced-parameter pole-zero filter. Our derivation successfully extends the "embedding" approach of [1,2] to the case of an unequal number of poles and zeros. Our main emphasis is on applying the algorithm to reduce the implementation complexity of the long FIR feedforward and feedback filters of the MMSE-DFE encountered in digital subscriber loops. We also describe several other applications where the algorithm leads to a signicant reduction in implementation complexity.