Recent trends in high-speed computing have moved towards the use of networks of workstations as a cost-effective approach to parallel computing. One recently proposed solution involves the use of an existing network of workstation-class computers as a single multiprocessor, and much research is ongoing in this area;This dissertation describes work in the area of process scheduling on networks of workstations, specifically in the area of load analysis. After presenting extensive background in the field, measures of CPU and network load are defined, and a "test" parallel application program presented, written for a network-multiprocessing software package called PVM. A series of experiments is then detailed, whose goal was to discover the relationship between the run time of the test application and the loads on the participating workstations and networks. The experiments include measurement of CPU loading and network loading, both during test application runs, during artificially elevated loads, and during "quiet" conditions. Results of the experiments are presented, and the applications of the results to the problem of task scheduling examined. It is then claimed that several easily measured load measures are useful to task scheduling, by allowing run time to be predicted within a margin of error, and allowing limiting network segments to be detected and avoided.