According to the present invention, there is provided a method of encoding a three-dimensional (3D) image, the method comprising: determining a prediction mode for a current block as an inter prediction mode; determining whether a reference block corresponding to the current block in a reference picture has motion information; when the reference block has the motion information, deriving motion information on the current block for each sub prediction block in the current block; and deriving a prediction sample for the current block based on the motion information on the current block.