The manual interpretation of a seismic horizon is a time consuming process, which drives the research for automatic or semi automatic tracking methods. Among the known propositions that use correlation, there is a common limitation: the usage of local approaches to determine which samples belong to the horizon. This kind of approach performs well in data where there are no seismi faults. However, by using only local information, it is prone to error propagation in low coherency areas, which usualy corresponds to fault regions. The goal of this work is to evaluate the performance of shortest path algorithms as a solution for the horizont tracking problem. It intends to propose a global method that is robust to different seismic features.