This paper investigates the principles of mapping linear recurrences on static data flow computers. For a linear recurrence with a single variable, the key is to properly introduce a feedback loop in the machine level data flow graphs. We show that, in order to achieve maximum pipelining, the critical dependence delay of the recurrence must be matched with the necessary computational delay of the graph. Two possible mapping techniques are discussed, which are (1) changing the dependence delay by introducing an additional companion pipeline; (2) changing the computational delay by inserting FIFOs. The mapping of the Val for-iter construct, the major language feature for expressing recurrences in Val, is outlined.