An oil pressure learning method of an automatic transmission includes acquiring a state of a vehicle in a state where relational regulation data is stored, supplying oil to the automatic transmission such that the value of the oil pressure is set to an oil pressure command value, calculating, as a specific variable, a variable representing an amount in which a detected input rotation speed exceeds a target input rotation speed, or the like, calculating a reward in a manner in which it has a larger value when the specific variable satisfies a criterion than when it does not satisfy the criterion, updating the relational regulation data by inputting, to an update mapping, the reward and the oil pressure command value, and calculating a torque variable having a value that is increased as an amount of change in an input torque is increased.