With the development of the high-speed communication channel or high-capacity storage device, the intersymbol interference (ISI) occurs more frequently in the condensed data. The traditional Bahl -Cocke -Jelineu -Raviv (BCJR) method is capable of solving the two-dimensional (2D) symbol influence problem with extremely high cost. By analysing the complexity in terms of branch, state and path in the 2D trellis-based ISI detection, the authors verify that the system complexity is mainly dominated by the window size. To reduce the detection complexity, the conventional simplifications are applied in the 2D stripwise ISI detection. In addition, a two-phase strip-wise detection is proposed to achieve a lower complexity than that of the conventional simplifications. In this system, the first detector can be realised by a low-complexity method (like the hard detection or Viterbi algorithm) to provide the mask shrinking with predicted data. After reducing the mask size, the second detector can detect the main portion of ISI by the trellis-based detection (like BCJR algorithm) with a small mask. Under the similar detection performance, the proposed scheme can achieve more than 91.76% saving in the metric computation compared with the 5 × 3-window IRCSDFA (iterative row-column soft-decision feedback algorithm) simplified by the M-algorithm with M = 8.