TAR-Crash (TARC) is developed aiming at the limitations of current software. First of all, different vehicles types and accident types are considered to enhance its applicability, among which vehicle to pole/tree and vehicle-pedestrian crash accident are deeply investigated. Crash models based on improved momentum conservation law are established according to the above types. Critical coefficients in the model are calibrated from real-crash test data. Then, models are coded and programmed based on Visual C++. Finally, three real-world accidents involving different vehicle accident types are reconstructed in TARC. Results show the reliability, accuracy and convenience of this newly developed software.