Nowadays many researches have focused on structural ET based on statement and branch coverage and there are few researches on path-oriented ET. To solve this problem, this paper provokes an approach to construct the fitness function for test case generation in path-oriented ET based on the similarity evaluation techniques. First, a basic model for fitness function design is provided. The core of the model is to evaluate the similarity between the execution track and the target path. Accordingly three different algorithms for the similarity evaluation are provided. This model can automatically generate fitness function for each target path. The empirical studies present the superiority of the approach over several other path-oriented testing techniques, especially for the complex paths. Besides, the limitation and the applicable scope of the approach are pointed out.