The PARKBENCH kernel benchmarks are written as far as possible in standard Fortran 77 using 64-bit floating point arithmetic ( DOUBLE PRECISION on most systems), unless otherwise stated. Both PVM/MPI [2,3] and subset HPF versions exist for most of the codes in addition to the standard Fortran-77 versions. A description of each benchmark and instructions on how to run it are given in individual ReadMe files. They also contain a specification of the three problem sizes agreed upon for each code: (1) test problem (2) moderate size and (3) grand challenge size. A formula should be given in the ReadMe files to produce flop counts for the kernel benchmarks along with precalculated figures for each standard problem size. Make-files are supplied with each benchmark to handle compilation and linking in a Unix environment.