The Worst-Case Execution Time (WCET) of real-time systems is mainly influenced by the program design, its execution environment and the input data. To cover the last factor in the context of WCET estimation, the objective of this work is to generate the test-data that maximize the execution times of the parallel real-time systems. In this paper, a test-data generation technique is proposed that uses Genetic Algorithms to automatically generate the input data, to be used for testing of parallel real-time systems. The proposed technique was applied to a parallel embedded application – Stringsearch. The result was an analysis that took as input the parallel program and generated the test-data that cause maximal execution times. The generated test-data showed improvements by exercising long execution times in comparison to randomly generated input data.