The MARS15 Monte Carlo code capabilities to deal with time-consuming deep penetration shielding problems and other computationally tough tasks in accelerator, detector and shielding applications, have been enhanced by a parallel processing option. It has been developed, implemented and tested on the Fermilab Accelerator Division Linux cluster and network of Sun workstations. The code uses a message passing interface MPI. It is scalable and demonstrates good performance. The general architecture of the code, specific uses of message passing and effects of a scheduling on the performance and fault tolerance are described.