This paper proposes a class of graph rules for deducing associations between entities, referred to as Graph Rules with Oracles and denoted by GROs. As opposed to previous graph rules, GROs support oracle functions to import (a) external knowledge, and (b) internal computations such as aggregate operators and machine learning predicates, and so on. Moreover, the semantics of GROs are defined in terms of pivoted dual simulation, in contrast to the subgraph isomorphism. We show how GROs can be used to predict links and catch anomalies, among other things. We formalize the association deduction problem with GROs in terms of the chase, and prove their Church-Rosser property. We show that both the deduction and incremental deduction problems with GROs are in PTIME, as opposed to the intractability of their counterparts with prior graph rules. We also provide sequential and parallel algorithms for association deduction and incremental deduction. Using real-life and synthetic graphs, we experimentally verify the effectiveness, scalability, and efficiency of the algorithms.