Anthropomorphic test devices (ATDs) are useful for simulating human damage during traffic accidents. For large accidents such as train accidents, experiments using full-size ATDs are not feasible because of cost, space, and time requirements. However, it is difficult to develop small-scale ATDs because changes in geometry are not necessarily accompanied by commensurate changes in mass or force. In this study, we designed and developed small-scale human dummies using the similarity rule. We determined the similarity ratio and focused on the neck, chest, and abdomen, which are important for protecting organs and nerves. Drop, impact, and sled tests based on ISO TR9790 indicators were used to evaluate the ATD. The first ATD scored 4.69, which indicates 'fair biofidelity'; the neck and chest delivered low scores during the sled and impact tests, respectively. We simulated the behavior of the ATD using finite element analysis; the experimental and analytical values were consistent. We modified the neck and chest parameters using simulation results and evaluated the optimized ATD using impact and sled tests. The optimized ATD scored a 6.56, which indicates 'good biofidelity.' In conclusion, we developed a small-scale ATD capable of satisfactorily simulating human behavior. Using the proposed ATD, we can reduce the opportunities for full-scale experiments.