With the demand for sophisticated techniques to easily prevent the deflection of needles, robotic CT (computed tomography)-guided puncture with an ultrafine needle is being investigated. Quantification of deflection is essential for accurate puncture with ultrafine needles. Research on the quantification of deflection caused by tissue reaction forces to which the beveled surface of the needle tip is in progress, and this method has been applied for deflection reduction and needle steering within the tissue. However, when the needle tip passes through a tissue boundary, the needle deflects regardless of the direction of the beveled surface. Although several methods have been proposed to reduce the deflection caused by the boundary surface, no curve puncture method has been constructed using the deflection. This work aimed to construct a refraction model that can back-calculate the curved path from the body surface to the target. Assumptions of the refraction model were made based on the results of ex vivo examination, and the model was validated through in vivo examination. A refraction model in which the refraction angle is linearly proportional to the needle penetration angle relative to the boundary was hypothesized. Validation test revealed that the correlation coefficient exceeded 0.9, which was similar to that of the model and suggested the biological adaptability of the proposed model. A curve puncture method using this refraction model will be developed in the future.