Knowledge is a critical property that organizations use to gain and maintain competitive advantages. In the constantly changing business environment, organizations have to exploit effective and efficient approaches to help knowledge workers find task-relevant knowledge, as well as to preserve, share and reuse such knowledge. Hence, an important issue is how to discover knowledge flow (KF) from the historical work records of knowledge workers in order to understand their task-needs and the ways they reference documents, and actively provide adaptive knowledge support. This work proposes a KF-based document recommendation method that integrates KF mining and collaborative filtering recommendation mechanisms to recommend codified knowledge. The approach consists of two phases: the KF mining phase and the recommendation phase. The KF mining phase can identify each worker's knowledge flow by considering the referencing time and citation relations of knowledge resources. Then, based on the discovered KF, the recommendation phase applies sequential rule mining and the CF method to recommend relevant documents to the target worker. Experiments are conducted to evaluate the performance of the proposed method and compare it with the traditional CF method using data collected from a research institute laboratory. The experiment results show that the proposed method can improve the quality of recommendation.