The increasing penetration of the real world with embedded and globally networked sensors enables the formation of a Web of Things (WoT), where high-level state information derived from sensors is embedded into Web representations of real-world entities (e.g. places, objects). A key service for the WoT is searching for entities which exhibit a certain dynamic state at the time of the query, which is a challenging problem due to the dynamic nature of the sought state information and due to the potentially huge scale of the WoT. In this paper we introduce a primitive called sensor ranking to enable efficient search for sensors that have a certain output state at the time of the query. The key idea is to efficiently compute for each sensor an estimate of the probability that it matches the query and process sensors in the order of decreasing probability, such that effort is first spent on sensors that are very likely to actually match the query. Using real data sets, we show that sensor ranking can significantly improve the efficiency of content-based sensor search.