The present disclosure provides is a high-speed image recognition method and apparatus using a 3D CNN. The high-speed image recognition method using the 3D CNN includes: inputting each of a first image clips among image clips constituting an input image to the 3D CNN; acquiring softmax function values calculated in the 3D CNN with respect to each of the first image clips; calculating a score margin by using the softmax function values; comparing the score margin with a predetermined threshold value to determine whether to input at least one additional image clip other than the first image clips among the image clips constituting the input image to the 3D CNN. Therefore, a calculation speed for image recognition can be improved.