Model-predictive direct-power control (MPDPC) has attracted increasing attention from scholars because of its outstanding dynamic response and high-power factor. However, its major shortcoming is the high-computational cost, which limits its application in three-level (3L) converters. By dividing the conventional MPDPC into two steps, a double-step MPDPC (DMPDPC) algorithm is presented to address this problem. The first step of this algorithm is dedicated to searching a sector including an optimal switching state (OSS). In the second step, the OSS in a chosen sector is gained. Furthermore, its driver signals are changed as soon as the OSS is gained and their delay time is detected. Then, the corresponding delay compensation is also proposed to improve the steady-state performance of the 3L-converter. The validity of the proposed approach is demonstrated by the experimental platform with 4 kVA rated power. The experimental results show that the computing time in DMPDPC is 50.82% of that in conventional MPDPC. In addition, the proposed DMPDPC decreases the 3L converter's active and reactive powers ripples, reduces its current harmonic and strengthens its robustness against parameter variation.