A prediction control of a process containing a non-linear behavior is effected by predicting a transition of process quantities (control quantities) of an object to be controlled a predetermined time period after the current time to provide predicted values and by determining manipulation quantities of at least a control effector in accordance with a difference between the predicted values and predetermined target values of the controlled object. A quantitative operation is carried out arithmetically identifying a process as a linear behavior, and a fuzzy inference or qualitative operation, including a fuzzy rule based on empirical knowledge, is carried out for simulating a process, wherein for input process information and current manipulation quantities to be maintained, both operations parallelly determine predicted values of control quantities. A process behavior determination is arithmetically carried out to determine whether the process linearly behaves (i.e. the normal state or the state dominant to statistic distribution) or not (i.e. the extreme transition state such as a transient response, or an indeterminate state) from the specified process information on the basis of a process behavior determination rule. As a result, a predicted value acquired by the quantitative operation is selected in response to a determination of a linear behavior, and a predicted value acquired by the fuzzy inference operation is selected in response to a determination of a non-linear behavior.