The LLILOG knowledge representation language and an inference engine to interpret it have been developed as part of the LILOG project, where new concepts for understanding natural-language texts were investigated. LLILOG is a typed predicate logic whose type system has adopted the concepts of KL-One-like languages. Further language constructs allow the formulation of default and control knowledge. The inference engine for LLILOG was designed as an experimental theorem prover, allowing us to investigate the behavior of various inference calculi as well as a number of search strategies. Processing with LLILOG is not restricted to a propositional reasoner for logical formulas; we are also able to delegate special kinds of inferences to external deductive components. Currently, one such external reasoner for processing spatial information on the basis of analog representation is attached to the inference engine.