Inference in rule-based systems is based on logic. The scheme of rules is similar to implication in syntax and in semantics.
Two basic directions of inference are theoretically possible: forward chaining, or data-driven inference and backward-chaining, or goal directed inference.