In this paper we present a Linda implementation on a distributed memory parallel computer. Linda is a parallel programming paradigm providing a small set of extensions to any of the sequential languages such as C, FORTRAN: : : . This paradigm allows the porting of programs between various architectures of parallel computers. It is based on a shared associative memory, namely the tuple space. On distributed memory systems, this shared memory has to be simulated. Our implementation runs on a transputer network with any underlying topology. We first present the Linda operators, then the protocol used to manage the tuple space is described, and a dynamic load balancing scheme is also presented.