An ecosystem shifts to an alternative stable state when a threshold of accumulated pressure (i.e. direct impact of environmental change or human activities) is exceeded. Detecting this threshold in empirical data remains a challenge because ecosystems are governed by complex interlinkages and feedback loops between their components and pressures. In addition, multiple feedback mechanisms exist that can make an ecosystem resilient to state shifts. Therefore, unless a broad ecological perspective is used to detect state shifts, it remains questionable to what extent current detection methods really capture ecosystem state shifts and whether inferences made from smaller scale analyses can be implemented into ecosystem management. We reviewed the techniques currently used for retrospective detection of state shifts detection from empirical data. We show that most techniques are not suitable for taking a broad ecosystem perspective because approximately 85% do not combine intervariable non‐linear relationships and high‐dimensional data from multiple ecosystem variables, but rather tend to focus on one subsystem of the ecosystem. Thus, our perception of state shifts may be limited by methods that are often used on smaller data sets, unrepresentative of whole ecosystems. By reviewing the characteristics, advantages, and limitations of the current techniques, we identify methods that provide the potential to incorporate a broad ecosystem‐based approach. We therefore provide perspectives into developing techniques better suited for detecting ecosystem state shifts that incorporate intervariable interactions and high‐dimensionality data.