A method of finding the prediction direction during intraframe video image coding. The method comprises the following steps: selecting two initial prediction directions (E, S), calculating virtual blocks associated with said initial prediction directions (E, S), comparing said virtual blocks with the current block and selecting the initial prediction direction (E), which is then called the first main direction, calculating virtual blocks associated with the prediction directions (ENE, ESE) immediately adjacent to said first main direction (E), comparing said virtual blocks with the current block to determine the best prediction direction, if the best prediction direction is the first main direction (E) or one direction (ENE) of the immediately adjacent directions situated at one extremity of the set of prediction directions, deciding this best direction is the required prediction direction, otherwise, calculating the virtual block associated with the prediction direction (SE) other than the first main direction (E) that is immediately adjacent to the second main direction (ESE), comparing said virtual blocks with the current block to determine the best prediction direction, if said best prediction direction is said second main direction (ESE) or the immediately adjacent direction (SE) situated at one extremity of the set of prediction directions, deciding that best direction is the required prediction direction, otherwise, continuing the process iteratively until the required prediction direction is found. Application to video coding.