In this paper we describe the issues in the design of CHRONUS - a spontaneous speech understanding system. The CHRONUS system is based on the approach we proposed in 1991, which formalizes the understanding problem as a communication problem. The model assumes that a spoken sentence is generated by a HMM like process whose hidden states correspond to elemental meaning units called concepts. Understanding therefore consists in decoding the hidden concepts given a spoken utterance [2]. The CHRONUS system was used in the ARPA ATIS task, a spontaneous speech database interface. In the 1994 official evaluation it achieved a state of the art accuracy. In the following paper we discuss the features of the system that contributed to this success.