An apparatus of controlling a safety power window of a vehicle and a method thereof to not injure the body caught in a window frame, as well as to prevent a malfunction of the safety power window, by adaptively setting an operation reference value of the safety power window, may include a learning device that deep-learns a model which predicts an operation reference value of the safety power window provided in the vehicle, based on driving information of the vehicle, a sensor that collects the driving information of the vehicle, and a controller that obtains the operation reference value of the safety power window corresponding to current driving information of the vehicle by use of the model on which the deep learning is completed, and controls the safety power window according to the operation reference value of the safety power window.