Many ITS applications rely on known attributes of the traffic conditions. One useful property is congestion state which allows for differential behaviour in the system when demand is below, at or above capacity. Congestion detection in certain common data types such as loop detectors is frequently and idiosyncratically addressed by many researchers and practitioners. A set of flexible, objective and robust methods would facilitate the comparison of congestion state across datasets, locations and times of day to better model the response of the system to ITS interventions. This work develops geometric congestion detection algorithms for use in speed-flow and flow-density space. The methods are applicable to any dataset comprised of vehicle flows and speeds (such as loop detector data). The speed-flow space algorithm attempts to identify clusters in speed-flow space based on effective capacity and a cut-off free flow speed. The flow-density diagram builds on the theory supporting the triangular fundamental diagram and classifies congestion based on a density cut-off. Both methods incorporate time-of-day selection. The methods are successful in identifying clearly congested or uncongested observations along a test corridor. In conjunction, the two methods are able to distinguish two regions of ambiguity associated with the transition from uncongested to congested and vice versa. The combination of the two methods offers a promising approach for quickly and robustly classifying observations from a variety of location-typologies into two, three or four traffic states depending on the application.