A conference key distribution is a protocol that allows designated subset of users to calculate a shared private key. We consider unconditionally secure conference key distribution systems where the adversary has unlimited computational power, and focuses on a stronger and more realistic adversary model, proposed by Safavi-Naini and Jiang, in which the adversary in addition to corrupting subsets of users and obtaining their private keys, can access the conference keys of a number of uncorrupted conferences. We consider alternative definitions of security with this adversary model and show the relationship between them. An important efficiency parameter for conference key distribution systems is the size of the users' private keys. We derive lower bounds on the size of this key and discuss the results in comparison with the known bounds. We also consider one-round Interactive Conference Key Distribution Systems (ICKDS) in the new adversarial model where the adversary in addition to learning the private keys of the corrupted users and a number of conference keys, has access to the transcripts of some conferences. We show that under this new adversary model, the previously proposed one round protocol of Blundo et. al. is no longer secure and we construct an ICKDS with provable security in the new adversarial model.