Liquefaction is one of the major causes of ground failures during an earthquake. Recent evidence shows that the existing variants of the "simplified" liquefaction evaluation procedure lead to inaccurate results for megathrust earthquakes in subduction interfaces. To overcome this drawback and to achieve better prediction of liquefaction cases in subduction zones, this research intends to develop new empirical models that could be used for the prediction of liquefaction triggering in subduction zones. Towards this goal, new models for number of equivalent cycles (n_eq) and stress-reduction factor (r_d) have been proposed. The models are developed by regressing site response data obtained from 254 pairs of subduction ground motions and 77 representative soil profiles. To account for tectonic differences and magnitude scaling, separate models are developed for interface and intraslab earthquakes. The uncertainties involved in the proposed models are quantified through standard deviations of regression coefficients, event, site, and residual terms. The resulting models differ from other published models, especially the model for number of equivalent cycles. It was found that n_eq is greatly influenced by the fundamental site period. The model for r_d predicts higher values at shallow depths and lower values at deeper layers than other published models. Comparing the factors of safety against liquefaction with those from other existing models revealed that the use of models proposed in this research is more likely to reduce the "false positives" in liquefaction predictions, especially when design ground motion acceleration is high. / Master of Science / During earthquake shaking, loose saturated sands may lose strength and behave more like a liquid than a solid. This phenomenon is referred to liquefaction. Liquefaction has been responsible for infrastructure failure during past earthquakes, thus leading to major economic losses. This prompts the prediction and mitigation of potential liquefaction effects in a building site. However, the current state-of-the-practice for predicting liquefaction is inaccurate for large magnitude earthquakes in subduction zones. This provided the impetus for this research which focusses on developing new liquefaction evaluation models for large magnitude earthquakes. New models for number of equivalent cycles and stress reduction factor are developed by analyzing the representative ground motions and soil strata. These empirical parameters are central to the prediction of liquefaction triggering. Comparing the new models with the existing models revealed that the factor of safety against liquefaction estimated using new models are greater than those obtained using existing models for large magnitude earthquake scenario when the ground acceleration is high. This implies that using the existing models for predicting liquefaction in a site subjected to high values of ground acceleration from a subduction earthquake will lead to "false positives." Developed using a comprehensive dataset and robust regression techniques, the models developed in this research will lead to better predictions of liquefaction due to large subduction events.