Jade is a language designed to support coarse-grain parallelism on both shared and distributed address-space machines. Jade is data-oriented: a Jade programmer simply augments a sequential imperative program with declarations specifying how the program accesses data. A Jade implementation dynamically interprets the access specification to execute the program concurrently while enforcing the program's data dependence constraints, thus preserving the sequential semantics.This paper describes the Jade constructs and defines both a serial and a parallel formal operational semantics for Jade. The paper proves that the two semantics are equivalent.