Abstract: Process models are becoming more and more widespread in contemporary organizations. It is a complex and high-cost work to develop individual process models for specific business requirements. The modeling procedure can be accelerated and cost-decreased by using reference process models as a basis for individual process models development, so reference process models are widely adopted by organizations. Because building reference process models requires a mass of modeling and analyzing work by domain experts, a major challenge has emerged that how to automatically build a preliminary reference model inductively based on the existing process variants to provide assistance to domain experts. The existing methods of building reference process models have some shortcomings, such as output reference models of most methods have high model complexity and reference models described by traditional process modeling language could not entirely represent various recommended practice in a specific domain. To build reference process models with high representativeness and understandability, this paper proposes an approach to automatically build customizable reference models which support hierarchical sub-process based on fragments clustering. The base model, change options and constraints in customized process models are fully supported to build automatically by our method. The evaluation results show that the generated reference models could achieve fine domain representativeness and model complexity.