ABSTRACT The vulnerability assessment of power systems consists on finding the set of most critical assets in order to device strategies to make the system more resilient. The electric grid interdiction problem (EGIP), also known as the terrorist threat problem, addresses this issue by modeling the interaction of a disruptive agent and the system operator. The EGIP is usually modeled as a bilevel programming problem. The disruptive agent is placed in the upper-level optimization problem and aims at maximizing the system damage subject to limited destructive resources. The system operator is placed in the lower-level optimization problem and reacts to the attacks minimizing load shedding by redispatching available generation resources. Traditional approaches to the EGIP consider a simplified version of the network by means of a DC model. This allows some advantages from the standpoint of complexity; nevertheless, the effect of reactive power and voltage magnitudes are neglected in this model. An AC modeling of the network is more accurate but implies higher complexity. This paper presents a comparison of these models applied to the EGIP through an Iterated Local Search metaheuristic. Several tests were performed on a benchmark power system to contrast the performance of both models. Results show that using a DC model provides faster results but also reports conservative solutions that do not fully take into account the actual damage inflicted in the network. This might lead the system operator to underestimate the real vulnerably of the system and not carry out effective corrective or protective actions.