An image decoding method decodes a coded stream which includes plural processing units and a header of the processing units, and which is generated by coding a moving picture using inter prediction, the processing units including at least one processing unit divided in a hierarchy, the hierarchy including: a highest hierarchical layer in which a coding unit exists as a largest processing unit; and a lower hierarchical layer in which a prediction unit exists. The method comprises: identifying, by parsing hierarchy depth information stored in the header and indicating a hierarchical layer higher than a lowest hierarchical layer in which a smallest prediction unit exits, a hierarchical layer which is indicated by the hierarchy depth information or a hierarchical layer higher than the indicated hierarchical layer, the hierarchical layer having a prediction unit in which a reference index is stored; and decoding the prediction unit, using the reference index.