To develop a function generating method and a function generating device with which an error can be reduced without increasing the number of sample points to be stored in a memory.

In a function generating method and its device, an argument is discretized into Q (Q is a natural number) points, a function value to a reference argument obtained by discretion is made a reference function value, an interpolation coefficient accompanying it is stored in the memory, a function value and a correction coefficient to be references are read from the memory using a high-order bit of an argument provided in binary digit as an input address, the approximate value of the function value is calculated by linear interpolation from an argument for interpolation operation obtained from a low-order bit of the argument and each correction coefficient, further linear interpolation to the approximate value is performed from the argument for interpolation operation obtained from the low-order bit of the argument and another correction coefficient read from the memory, and the approximate value of the function is obtained by repeating the linear interpolation and making it multiple linear approximation.