PURPOSE: To prevent a learning region from becoming unstable by learning deflection from the standard value of air fuel ratio feedback control value in the operational state which includes an intake air quantity and the position of a gear as its parameters.

CONSTITUTION: In a controller 3, the learning control of air fuel ratio is performed to absorb variations in characteristics of an injector 4 and an intake air quantity sensor 2. Here, a learning region is divided, by using the intake air quantity as a parameter in each gear position, and division of an operational state region by the parameter of the intake air quantity is made different depending upon each gear position. Thus, every learning value can be fully learned and renewed, and therefore, unstability of the learning region can be prevented.