The computer systems of the eighties are expected to be designed using powerful low-cost distributed parts to achieve increases in computing power and concurrency. The resulting complexity due to the interaction and communication between these parts requires new methods for the analysis of the behavior of these systems. One such class of architectures, based on the concept of data flow, is designed to exploit the inherent parallelism within a program. In these computers, traditional sequencing constraints are removed and an operation is enabled for execution as soon as its operands are available;A method for approximating the time required to execute a data flow program (assuming adequate computing resources) is described. This method is applied to the static program graph at compile time and yields a parameterized equation for execution time performance. Based on a Petri net analysis and combined with more traditional approaches, this method is recursively applied to abstract operations in the program graph using a top-down approach. While this approach may introduce approximations at each stage, the major benefit is a significant reduction over other techniques in the time required for the analysis.