Today it is possible to deploy sensor networks in the real world and collect 
large amounts of raw sensory data. However, it remains a major challenge to 
make sense of sensor data, i.e., to extract high-level knowledge from the raw 
data. In this paper we present a novel in-network knowledge discovery 
technique, where high-level information is inferred from raw sensor data 
directly on the sensor nodes. In particular, our approach supports the 
discovery of frequent distributed event patterns, which characterize the 
spatial and temporal correlations between events observed by sensor nodes in a 
confined network neighborhood. One of the key challenges in realizing such a 
system are the constrained resources of sensor nodes. To this end, our 
solution offers a declarative query language that allows to trade off detail 
and scope of the sought patterns for resource consumption. We implement our 
proposal on real hardware and evaluate the trade-off between scope of the 
query and resource consumption.