In traditional security systems, for each task, we either trust an agent or we don't. If we trust an agent, we allow this agent full access to this particular task. This agent can usually allow his trusted sub-agents the same access, etc. If a trust management system only uses "trust" and "no trust" options, then a person should trust everyone in this potentially long chain. The problem is that trust is rarely a complete trust, there is a certain probability of distrust. So, when the chain becomes long, the probability of a security leak increases. It is desirable to keep track of trust probabilities, so that we should only delegate to agents whose trust is above a certain threshold. In this paper, we present efficient algorithms for handling such probabilities.