This paper presents a generalization of cubic phase function (CPF) for two-dimensional (2-D) cubic phase polynomial phase signals (PPS). Since a straightforward application of the CPF to the 2-D PPS leads to a demanding three-dimensional (3-D) search an efficient implementation is proposed by using genetic algorithms. Simulation results demonstrate that the proposed approach outperforms the classical Francos–Friedlander technique in terms of lower SNR threshold.