In the work of Sapthotharan K. Nair and J. Moon (IEEE Trans. Neural Nets., vol. 8, no. 5, pp. 1037-1048, 1997), the linear equalizer in PRML was replaced with a partial response neural equalizer (PRNEML) to combat nonlinear distortions, but overall complexities remained very high. Considering the RLL d=2 constraint, the Viterbi detector can be replaced with a simple discrete filter P(D-1) matched to target response P(D) (S. Gopalaswamy and R. Wood, Proc. SPIE vol. 3109, pp. 95-97, 1997). By using P(D)=1+D+D2 as a target response, discrete matched filter P(D-1) has no multiplications. Another advantage of the discrete matched filter structure is that decision feedback can be incorporated into neural networks. Overall structure of the decision feedback neural equalizer with discrete matched filter (DFNE/DMF) is shown. An optical recording channel is considered here, but the proposed system can be also applied to nonlinear magneto-optical and magnetic recording channels.