A learning device learns a control parameter (e.g., injection start response delay), which is used for deciding a control content of an injector (controlled object), in relation to a criterion variable (e.g., fuel pressure). The learning device has a storing section for storing a learning vector consisting of the control parameter and the criterion variable. The learning device has a measurement vector obtaining section for obtaining a measurement vector consisting of a measurement value of the control parameter and a measurement value of the criterion variable. The learning device has a correcting section for correcting the learning vector based on the measurement vector and for performing storing and updating of the learning vector in the storing section.