Fisheries are complex systems affected by environmental variability, species interactions, and human behavior. As such, this dissertation aims to study them as social-ecological systems. First, by questioning current modeling approaches, and then, by proposing the use of new methods that account for their inherent complexities. Chapter 1 asks whether aiming for Maximum Sustainable Yield, a standard objective for sustainable exploitation, could also generate economic stability for fishing communities in the Gulf of California, Mexico. We found that sustainable practices could increase total revenues by more than 70%, however, this would not be enough for 80% of fishers in the region to live above local poverty levels. As such, I frame the possibility to move away from traditional, equation-based, fisheries management towards dynamic and adaptive frameworks. Chapter 2 explores the use of Empirical Dynamics Modeling (EDM), a nonlinear and nonparametric method, to study marine ecosystems. By using a long-term planktonic time series from the North Sea, we found that longer time series help to detect nonlinear and state-dependent processes, also improving time series’ predictability. Chapter 3 uses a global database of stock assessments to find that traditional stock-recruitment models are somewhat successful at predicting data derived from assessment methods that introduce assumed stock-recruitment relationships. However, they are poor at predicting data that does not make such assumptions. We demonstrate that EDM is a better framework to predict future recruitment overall. Chapter 4 uses EDM to find that environmental processes and fishing pressure have both a detectable and comparable effect on the Pacific sardine’s population dynamics in the Gulf of California, traditionally thought to be affected only by long-term climatic variability. We develop an EDM-based model using fishing and environmental effects to predict catch two years ahead. We then use these predictions to propose an exploitation scheme that challenges the current policy that sets a constant harvest rate. This dissertation questions the use of equation-based models for fisheries management. Instead, it proposes the use of EDM as a way not only to improve real-world predictability, but also to consider both ecological and social processes with a unified quantitative approach.