The leak localization (LL) functionality is an important piece of leak detection systems that continuously monitor the operation of pipelines in real-time. After a leak is detected, it must be located as soon as possible so that mitigating actions can be taken to remedy the undesirable hazardous leak consequences. The most popular methodology used to locate leaks in single-phase pipelines is based on the pressure-gradient along the line. This technique uses the SCADA data polled and collected at the ends of the line segment, along with a fluid dynamics mechanical model capable of describing the flow in steady-state, to infer the leak position. By using the flow data at the pipe ends, the LL methodology seeks the interception of the pressure gradients computed by marching forward and backward along the line to estimate the leak site. This paper presents a detailed study to extend the LL strategy based on the pressure-gradient to two-phase flows in gas pipelines. To achieve this goal, without appealing to field or experimental data, we have built a leak database for typical real-world two-phase gas pipelines under the stratified-pattern flow by means of numerical simulations. By using a transient four-equation single-pressure two-fluid model, with suitable source terms to account for the leak existence, numerical simulations emulating leaks in a pipeline were carried out using the Flux-Corrected Transport method. After being validated by the commercial simulator OLGA, the mathematical model was employed to generate a leak database contemplating leaks of three different magnitudes at three distinct locations along the line for two different scenarios considering low and high holdup contents. The performance of the LL strategy, which has used as mechanical model the steady-state parcel of the transient mathematical model described above along with the simulated data at the pipe inlet and outlet, was evaluated by computing the leak location relative errors. The obtained results have shown that the pressure-gradient methodology has presented a particularly good performance under the circumstances considered herein. It was able to locate leaks of magnitudes between 1% and 10% of the total mass flowrate at the pipe inlet with relative errors ranging from 0.26% to 10.0%.